1 // SPDX-License-Identifier: GPL-2.0-or-later 2 3 #include <linux/mrp_bridge.h> 4 #include "br_private_mrp.h" 5 6 static const u8 mrp_test_dmac[ETH_ALEN] = { 0x1, 0x15, 0x4e, 0x0, 0x0, 0x1 }; 7 static const u8 mrp_in_test_dmac[ETH_ALEN] = { 0x1, 0x15, 0x4e, 0x0, 0x0, 0x3 }; 8 9 static bool br_mrp_is_ring_port(struct net_bridge_port *p_port, 10 struct net_bridge_port *s_port, 11 struct net_bridge_port *port) 12 { 13 if (port == p_port || 14 port == s_port) 15 return true; 16 17 return false; 18 } 19 20 static bool br_mrp_is_in_port(struct net_bridge_port *i_port, 21 struct net_bridge_port *port) 22 { 23 if (port == i_port) 24 return true; 25 26 return false; 27 } 28 29 static struct net_bridge_port *br_mrp_get_port(struct net_bridge *br, 30 u32 ifindex) 31 { 32 struct net_bridge_port *res = NULL; 33 struct net_bridge_port *port; 34 35 list_for_each_entry(port, &br->port_list, list) { 36 if (port->dev->ifindex == ifindex) { 37 res = port; 38 break; 39 } 40 } 41 42 return res; 43 } 44 45 static struct br_mrp *br_mrp_find_id(struct net_bridge *br, u32 ring_id) 46 { 47 struct br_mrp *res = NULL; 48 struct br_mrp *mrp; 49 50 hlist_for_each_entry_rcu(mrp, &br->mrp_list, list, 51 lockdep_rtnl_is_held()) { 52 if (mrp->ring_id == ring_id) { 53 res = mrp; 54 break; 55 } 56 } 57 58 return res; 59 } 60 61 static struct br_mrp *br_mrp_find_in_id(struct net_bridge *br, u32 in_id) 62 { 63 struct br_mrp *res = NULL; 64 struct br_mrp *mrp; 65 66 hlist_for_each_entry_rcu(mrp, &br->mrp_list, list, 67 lockdep_rtnl_is_held()) { 68 if (mrp->in_id == in_id) { 69 res = mrp; 70 break; 71 } 72 } 73 74 return res; 75 } 76 77 static bool br_mrp_unique_ifindex(struct net_bridge *br, u32 ifindex) 78 { 79 struct br_mrp *mrp; 80 81 hlist_for_each_entry_rcu(mrp, &br->mrp_list, list, 82 lockdep_rtnl_is_held()) { 83 struct net_bridge_port *p; 84 85 p = rtnl_dereference(mrp->p_port); 86 if (p && p->dev->ifindex == ifindex) 87 return false; 88 89 p = rtnl_dereference(mrp->s_port); 90 if (p && p->dev->ifindex == ifindex) 91 return false; 92 93 p = rtnl_dereference(mrp->i_port); 94 if (p && p->dev->ifindex == ifindex) 95 return false; 96 } 97 98 return true; 99 } 100 101 static struct br_mrp *br_mrp_find_port(struct net_bridge *br, 102 struct net_bridge_port *p) 103 { 104 struct br_mrp *res = NULL; 105 struct br_mrp *mrp; 106 107 hlist_for_each_entry_rcu(mrp, &br->mrp_list, list, 108 lockdep_rtnl_is_held()) { 109 if (rcu_access_pointer(mrp->p_port) == p || 110 rcu_access_pointer(mrp->s_port) == p || 111 rcu_access_pointer(mrp->i_port) == p) { 112 res = mrp; 113 break; 114 } 115 } 116 117 return res; 118 } 119 120 static int br_mrp_next_seq(struct br_mrp *mrp) 121 { 122 mrp->seq_id++; 123 return mrp->seq_id; 124 } 125 126 static struct sk_buff *br_mrp_skb_alloc(struct net_bridge_port *p, 127 const u8 *src, const u8 *dst) 128 { 129 struct ethhdr *eth_hdr; 130 struct sk_buff *skb; 131 __be16 *version; 132 133 skb = dev_alloc_skb(MRP_MAX_FRAME_LENGTH); 134 if (!skb) 135 return NULL; 136 137 skb->dev = p->dev; 138 skb->protocol = htons(ETH_P_MRP); 139 skb->priority = MRP_FRAME_PRIO; 140 skb_reserve(skb, sizeof(*eth_hdr)); 141 142 eth_hdr = skb_push(skb, sizeof(*eth_hdr)); 143 ether_addr_copy(eth_hdr->h_dest, dst); 144 ether_addr_copy(eth_hdr->h_source, src); 145 eth_hdr->h_proto = htons(ETH_P_MRP); 146 147 version = skb_put(skb, sizeof(*version)); 148 *version = cpu_to_be16(MRP_VERSION); 149 150 return skb; 151 } 152 153 static void br_mrp_skb_tlv(struct sk_buff *skb, 154 enum br_mrp_tlv_header_type type, 155 u8 length) 156 { 157 struct br_mrp_tlv_hdr *hdr; 158 159 hdr = skb_put(skb, sizeof(*hdr)); 160 hdr->type = type; 161 hdr->length = length; 162 } 163 164 static void br_mrp_skb_common(struct sk_buff *skb, struct br_mrp *mrp) 165 { 166 struct br_mrp_common_hdr *hdr; 167 168 br_mrp_skb_tlv(skb, BR_MRP_TLV_HEADER_COMMON, sizeof(*hdr)); 169 170 hdr = skb_put(skb, sizeof(*hdr)); 171 hdr->seq_id = cpu_to_be16(br_mrp_next_seq(mrp)); 172 memset(hdr->domain, 0xff, MRP_DOMAIN_UUID_LENGTH); 173 } 174 175 static struct sk_buff *br_mrp_alloc_test_skb(struct br_mrp *mrp, 176 struct net_bridge_port *p, 177 enum br_mrp_port_role_type port_role) 178 { 179 struct br_mrp_ring_test_hdr *hdr = NULL; 180 struct sk_buff *skb = NULL; 181 182 if (!p) 183 return NULL; 184 185 skb = br_mrp_skb_alloc(p, p->dev->dev_addr, mrp_test_dmac); 186 if (!skb) 187 return NULL; 188 189 br_mrp_skb_tlv(skb, BR_MRP_TLV_HEADER_RING_TEST, sizeof(*hdr)); 190 hdr = skb_put(skb, sizeof(*hdr)); 191 192 hdr->prio = cpu_to_be16(mrp->prio); 193 ether_addr_copy(hdr->sa, p->br->dev->dev_addr); 194 hdr->port_role = cpu_to_be16(port_role); 195 hdr->state = cpu_to_be16(mrp->ring_state); 196 hdr->transitions = cpu_to_be16(mrp->ring_transitions); 197 hdr->timestamp = cpu_to_be32(jiffies_to_msecs(jiffies)); 198 199 br_mrp_skb_common(skb, mrp); 200 201 /* In case the node behaves as MRA then the Test frame needs to have 202 * an Option TLV which includes eventually a sub-option TLV that has 203 * the type AUTO_MGR 204 */ 205 if (mrp->ring_role == BR_MRP_RING_ROLE_MRA) { 206 struct br_mrp_sub_option1_hdr *sub_opt = NULL; 207 struct br_mrp_tlv_hdr *sub_tlv = NULL; 208 struct br_mrp_oui_hdr *oui = NULL; 209 u8 length; 210 211 length = sizeof(*sub_opt) + sizeof(*sub_tlv) + sizeof(*oui) + 212 MRP_OPT_PADDING; 213 br_mrp_skb_tlv(skb, BR_MRP_TLV_HEADER_OPTION, length); 214 215 oui = skb_put(skb, sizeof(*oui)); 216 memset(oui, 0x0, sizeof(*oui)); 217 sub_opt = skb_put(skb, sizeof(*sub_opt)); 218 memset(sub_opt, 0x0, sizeof(*sub_opt)); 219 220 /* 32 bit alligment shall be ensured therefore add 2 bytes */ 221 sub_tlv = skb_put_zero(skb, sizeof(*sub_tlv) + MRP_OPT_PADDING); 222 sub_tlv->type = BR_MRP_SUB_TLV_HEADER_TEST_AUTO_MGR; 223 } 224 225 br_mrp_skb_tlv(skb, BR_MRP_TLV_HEADER_END, 0x0); 226 227 return skb; 228 } 229 230 static struct sk_buff *br_mrp_alloc_in_test_skb(struct br_mrp *mrp, 231 struct net_bridge_port *p, 232 enum br_mrp_port_role_type port_role) 233 { 234 struct br_mrp_in_test_hdr *hdr = NULL; 235 struct sk_buff *skb = NULL; 236 237 if (!p) 238 return NULL; 239 240 skb = br_mrp_skb_alloc(p, p->dev->dev_addr, mrp_in_test_dmac); 241 if (!skb) 242 return NULL; 243 244 br_mrp_skb_tlv(skb, BR_MRP_TLV_HEADER_IN_TEST, sizeof(*hdr)); 245 hdr = skb_put(skb, sizeof(*hdr)); 246 247 hdr->id = cpu_to_be16(mrp->in_id); 248 ether_addr_copy(hdr->sa, p->br->dev->dev_addr); 249 hdr->port_role = cpu_to_be16(port_role); 250 hdr->state = cpu_to_be16(mrp->in_state); 251 hdr->transitions = cpu_to_be16(mrp->in_transitions); 252 hdr->timestamp = cpu_to_be32(jiffies_to_msecs(jiffies)); 253 254 br_mrp_skb_common(skb, mrp); 255 br_mrp_skb_tlv(skb, BR_MRP_TLV_HEADER_END, 0x0); 256 257 return skb; 258 } 259 260 /* This function is continuously called in the following cases: 261 * - when node role is MRM, in this case test_monitor is always set to false 262 * because it needs to notify the userspace that the ring is open and needs to 263 * send MRP_Test frames 264 * - when node role is MRA, there are 2 subcases: 265 * - when MRA behaves as MRM, in this case is similar with MRM role 266 * - when MRA behaves as MRC, in this case test_monitor is set to true, 267 * because it needs to detect when it stops seeing MRP_Test frames 268 * from MRM node but it doesn't need to send MRP_Test frames. 269 */ 270 static void br_mrp_test_work_expired(struct work_struct *work) 271 { 272 struct delayed_work *del_work = to_delayed_work(work); 273 struct br_mrp *mrp = container_of(del_work, struct br_mrp, test_work); 274 struct net_bridge_port *p; 275 bool notify_open = false; 276 struct sk_buff *skb; 277 278 if (time_before_eq(mrp->test_end, jiffies)) 279 return; 280 281 if (mrp->test_count_miss < mrp->test_max_miss) { 282 mrp->test_count_miss++; 283 } else { 284 /* Notify that the ring is open only if the ring state is 285 * closed, otherwise it would continue to notify at every 286 * interval. 287 * Also notify that the ring is open when the node has the 288 * role MRA and behaves as MRC. The reason is that the 289 * userspace needs to know when the MRM stopped sending 290 * MRP_Test frames so that the current node to try to take 291 * the role of a MRM. 292 */ 293 if (mrp->ring_state == BR_MRP_RING_STATE_CLOSED || 294 mrp->test_monitor) 295 notify_open = true; 296 } 297 298 rcu_read_lock(); 299 300 p = rcu_dereference(mrp->p_port); 301 if (p) { 302 if (!mrp->test_monitor) { 303 skb = br_mrp_alloc_test_skb(mrp, p, 304 BR_MRP_PORT_ROLE_PRIMARY); 305 if (!skb) 306 goto out; 307 308 skb_reset_network_header(skb); 309 dev_queue_xmit(skb); 310 } 311 312 if (notify_open && !mrp->ring_role_offloaded) 313 br_mrp_ring_port_open(p->dev, true); 314 } 315 316 p = rcu_dereference(mrp->s_port); 317 if (p) { 318 if (!mrp->test_monitor) { 319 skb = br_mrp_alloc_test_skb(mrp, p, 320 BR_MRP_PORT_ROLE_SECONDARY); 321 if (!skb) 322 goto out; 323 324 skb_reset_network_header(skb); 325 dev_queue_xmit(skb); 326 } 327 328 if (notify_open && !mrp->ring_role_offloaded) 329 br_mrp_ring_port_open(p->dev, true); 330 } 331 332 out: 333 rcu_read_unlock(); 334 335 queue_delayed_work(system_percpu_wq, &mrp->test_work, 336 usecs_to_jiffies(mrp->test_interval)); 337 } 338 339 /* This function is continuously called when the node has the interconnect role 340 * MIM. It would generate interconnect test frames and will send them on all 3 341 * ports. But will also check if it stop receiving interconnect test frames. 342 */ 343 static void br_mrp_in_test_work_expired(struct work_struct *work) 344 { 345 struct delayed_work *del_work = to_delayed_work(work); 346 struct br_mrp *mrp = container_of(del_work, struct br_mrp, in_test_work); 347 struct net_bridge_port *p; 348 bool notify_open = false; 349 struct sk_buff *skb; 350 351 if (time_before_eq(mrp->in_test_end, jiffies)) 352 return; 353 354 if (mrp->in_test_count_miss < mrp->in_test_max_miss) { 355 mrp->in_test_count_miss++; 356 } else { 357 /* Notify that the interconnect ring is open only if the 358 * interconnect ring state is closed, otherwise it would 359 * continue to notify at every interval. 360 */ 361 if (mrp->in_state == BR_MRP_IN_STATE_CLOSED) 362 notify_open = true; 363 } 364 365 rcu_read_lock(); 366 367 p = rcu_dereference(mrp->p_port); 368 if (p) { 369 skb = br_mrp_alloc_in_test_skb(mrp, p, 370 BR_MRP_PORT_ROLE_PRIMARY); 371 if (!skb) 372 goto out; 373 374 skb_reset_network_header(skb); 375 dev_queue_xmit(skb); 376 377 if (notify_open && !mrp->in_role_offloaded) 378 br_mrp_in_port_open(p->dev, true); 379 } 380 381 p = rcu_dereference(mrp->s_port); 382 if (p) { 383 skb = br_mrp_alloc_in_test_skb(mrp, p, 384 BR_MRP_PORT_ROLE_SECONDARY); 385 if (!skb) 386 goto out; 387 388 skb_reset_network_header(skb); 389 dev_queue_xmit(skb); 390 391 if (notify_open && !mrp->in_role_offloaded) 392 br_mrp_in_port_open(p->dev, true); 393 } 394 395 p = rcu_dereference(mrp->i_port); 396 if (p) { 397 skb = br_mrp_alloc_in_test_skb(mrp, p, 398 BR_MRP_PORT_ROLE_INTER); 399 if (!skb) 400 goto out; 401 402 skb_reset_network_header(skb); 403 dev_queue_xmit(skb); 404 405 if (notify_open && !mrp->in_role_offloaded) 406 br_mrp_in_port_open(p->dev, true); 407 } 408 409 out: 410 rcu_read_unlock(); 411 412 queue_delayed_work(system_percpu_wq, &mrp->in_test_work, 413 usecs_to_jiffies(mrp->in_test_interval)); 414 } 415 416 /* Deletes the MRP instance. 417 * note: called under rtnl_lock 418 */ 419 static void br_mrp_del_impl(struct net_bridge *br, struct br_mrp *mrp) 420 { 421 struct net_bridge_port *p; 422 u8 state; 423 424 /* Stop sending MRP_Test frames */ 425 cancel_delayed_work_sync(&mrp->test_work); 426 br_mrp_switchdev_send_ring_test(br, mrp, 0, 0, 0, 0); 427 428 /* Stop sending MRP_InTest frames if has an interconnect role */ 429 cancel_delayed_work_sync(&mrp->in_test_work); 430 br_mrp_switchdev_send_in_test(br, mrp, 0, 0, 0); 431 432 /* Disable the roles */ 433 br_mrp_switchdev_set_ring_role(br, mrp, BR_MRP_RING_ROLE_DISABLED); 434 p = rtnl_dereference(mrp->i_port); 435 if (p) 436 br_mrp_switchdev_set_in_role(br, mrp, mrp->in_id, mrp->ring_id, 437 BR_MRP_IN_ROLE_DISABLED); 438 439 br_mrp_switchdev_del(br, mrp); 440 441 /* Reset the ports */ 442 p = rtnl_dereference(mrp->p_port); 443 if (p) { 444 spin_lock_bh(&br->lock); 445 state = netif_running(br->dev) ? 446 BR_STATE_FORWARDING : BR_STATE_DISABLED; 447 p->state = state; 448 clear_bit(BR_MRP_AWARE_BIT, &p->flags); 449 spin_unlock_bh(&br->lock); 450 br_mrp_port_switchdev_set_state(p, state); 451 rcu_assign_pointer(mrp->p_port, NULL); 452 } 453 454 p = rtnl_dereference(mrp->s_port); 455 if (p) { 456 spin_lock_bh(&br->lock); 457 state = netif_running(br->dev) ? 458 BR_STATE_FORWARDING : BR_STATE_DISABLED; 459 p->state = state; 460 clear_bit(BR_MRP_AWARE_BIT, &p->flags); 461 spin_unlock_bh(&br->lock); 462 br_mrp_port_switchdev_set_state(p, state); 463 rcu_assign_pointer(mrp->s_port, NULL); 464 } 465 466 p = rtnl_dereference(mrp->i_port); 467 if (p) { 468 spin_lock_bh(&br->lock); 469 state = netif_running(br->dev) ? 470 BR_STATE_FORWARDING : BR_STATE_DISABLED; 471 p->state = state; 472 clear_bit(BR_MRP_AWARE_BIT, &p->flags); 473 spin_unlock_bh(&br->lock); 474 br_mrp_port_switchdev_set_state(p, state); 475 rcu_assign_pointer(mrp->i_port, NULL); 476 } 477 478 hlist_del_rcu(&mrp->list); 479 kfree_rcu(mrp, rcu); 480 481 if (hlist_empty(&br->mrp_list)) 482 br_opt_toggle(br, BROPT_MRP_ENABLED, false); 483 } 484 485 /* Adds a new MRP instance. 486 * note: called under rtnl_lock 487 */ 488 int br_mrp_add(struct net_bridge *br, struct br_mrp_instance *instance) 489 { 490 struct net_bridge_port *p; 491 struct br_mrp *mrp; 492 int err; 493 494 /* If the ring exists, it is not possible to create another one with the 495 * same ring_id 496 */ 497 mrp = br_mrp_find_id(br, instance->ring_id); 498 if (mrp) 499 return -EINVAL; 500 501 if (!br_mrp_get_port(br, instance->p_ifindex) || 502 !br_mrp_get_port(br, instance->s_ifindex)) 503 return -EINVAL; 504 505 /* It is not possible to have the same port part of multiple rings */ 506 if (!br_mrp_unique_ifindex(br, instance->p_ifindex) || 507 !br_mrp_unique_ifindex(br, instance->s_ifindex)) 508 return -EINVAL; 509 510 mrp = kzalloc_obj(*mrp); 511 if (!mrp) 512 return -ENOMEM; 513 514 mrp->ring_id = instance->ring_id; 515 mrp->prio = instance->prio; 516 517 p = br_mrp_get_port(br, instance->p_ifindex); 518 spin_lock_bh(&br->lock); 519 p->state = BR_STATE_FORWARDING; 520 set_bit(BR_MRP_AWARE_BIT, &p->flags); 521 spin_unlock_bh(&br->lock); 522 rcu_assign_pointer(mrp->p_port, p); 523 524 p = br_mrp_get_port(br, instance->s_ifindex); 525 spin_lock_bh(&br->lock); 526 p->state = BR_STATE_FORWARDING; 527 set_bit(BR_MRP_AWARE_BIT, &p->flags); 528 spin_unlock_bh(&br->lock); 529 rcu_assign_pointer(mrp->s_port, p); 530 531 if (hlist_empty(&br->mrp_list)) 532 br_opt_toggle(br, BROPT_MRP_ENABLED, true); 533 534 INIT_DELAYED_WORK(&mrp->test_work, br_mrp_test_work_expired); 535 INIT_DELAYED_WORK(&mrp->in_test_work, br_mrp_in_test_work_expired); 536 hlist_add_tail_rcu(&mrp->list, &br->mrp_list); 537 538 err = br_mrp_switchdev_add(br, mrp); 539 if (err) 540 goto delete_mrp; 541 542 return 0; 543 544 delete_mrp: 545 br_mrp_del_impl(br, mrp); 546 547 return err; 548 } 549 550 /* Deletes the MRP instance from which the port is part of 551 * note: called under rtnl_lock 552 */ 553 void br_mrp_port_del(struct net_bridge *br, struct net_bridge_port *p) 554 { 555 struct br_mrp *mrp = br_mrp_find_port(br, p); 556 557 /* If the port is not part of a MRP instance just bail out */ 558 if (!mrp) 559 return; 560 561 br_mrp_del_impl(br, mrp); 562 } 563 564 /* Deletes existing MRP instance based on ring_id 565 * note: called under rtnl_lock 566 */ 567 int br_mrp_del(struct net_bridge *br, struct br_mrp_instance *instance) 568 { 569 struct br_mrp *mrp = br_mrp_find_id(br, instance->ring_id); 570 571 if (!mrp) 572 return -EINVAL; 573 574 br_mrp_del_impl(br, mrp); 575 576 return 0; 577 } 578 579 /* Set port state, port state can be forwarding, blocked or disabled 580 * note: already called with rtnl_lock 581 */ 582 int br_mrp_set_port_state(struct net_bridge_port *p, 583 enum br_mrp_port_state_type state) 584 { 585 u32 port_state; 586 587 if (!p || !test_bit(BR_MRP_AWARE_BIT, &p->flags)) 588 return -EINVAL; 589 590 spin_lock_bh(&p->br->lock); 591 592 if (state == BR_MRP_PORT_STATE_FORWARDING) 593 port_state = BR_STATE_FORWARDING; 594 else 595 port_state = BR_STATE_BLOCKING; 596 597 p->state = port_state; 598 spin_unlock_bh(&p->br->lock); 599 600 br_mrp_port_switchdev_set_state(p, port_state); 601 602 return 0; 603 } 604 605 /* Set port role, port role can be primary or secondary 606 * note: already called with rtnl_lock 607 */ 608 int br_mrp_set_port_role(struct net_bridge_port *p, 609 enum br_mrp_port_role_type role) 610 { 611 struct br_mrp *mrp; 612 613 if (!p || !test_bit(BR_MRP_AWARE_BIT, &p->flags)) 614 return -EINVAL; 615 616 mrp = br_mrp_find_port(p->br, p); 617 618 if (!mrp) 619 return -EINVAL; 620 621 switch (role) { 622 case BR_MRP_PORT_ROLE_PRIMARY: 623 rcu_assign_pointer(mrp->p_port, p); 624 break; 625 case BR_MRP_PORT_ROLE_SECONDARY: 626 rcu_assign_pointer(mrp->s_port, p); 627 break; 628 default: 629 return -EINVAL; 630 } 631 632 br_mrp_port_switchdev_set_role(p, role); 633 634 return 0; 635 } 636 637 /* Set ring state, ring state can be only Open or Closed 638 * note: already called with rtnl_lock 639 */ 640 int br_mrp_set_ring_state(struct net_bridge *br, 641 struct br_mrp_ring_state *state) 642 { 643 struct br_mrp *mrp = br_mrp_find_id(br, state->ring_id); 644 645 if (!mrp) 646 return -EINVAL; 647 648 if (mrp->ring_state != state->ring_state) 649 mrp->ring_transitions++; 650 651 mrp->ring_state = state->ring_state; 652 653 br_mrp_switchdev_set_ring_state(br, mrp, state->ring_state); 654 655 return 0; 656 } 657 658 /* Set ring role, ring role can be only MRM(Media Redundancy Manager) or 659 * MRC(Media Redundancy Client). 660 * note: already called with rtnl_lock 661 */ 662 int br_mrp_set_ring_role(struct net_bridge *br, 663 struct br_mrp_ring_role *role) 664 { 665 struct br_mrp *mrp = br_mrp_find_id(br, role->ring_id); 666 enum br_mrp_hw_support support; 667 668 if (!mrp) 669 return -EINVAL; 670 671 mrp->ring_role = role->ring_role; 672 673 /* If there is an error just bailed out */ 674 support = br_mrp_switchdev_set_ring_role(br, mrp, role->ring_role); 675 if (support == BR_MRP_NONE) 676 return -EOPNOTSUPP; 677 678 /* Now detect if the HW actually applied the role or not. If the HW 679 * applied the role it means that the SW will not to do those operations 680 * anymore. For example if the role ir MRM then the HW will notify the 681 * SW when ring is open, but if the is not pushed to the HW the SW will 682 * need to detect when the ring is open 683 */ 684 mrp->ring_role_offloaded = support == BR_MRP_SW ? 0 : 1; 685 686 return 0; 687 } 688 689 /* Start to generate or monitor MRP test frames, the frames are generated by 690 * HW and if it fails, they are generated by the SW. 691 * note: already called with rtnl_lock 692 */ 693 int br_mrp_start_test(struct net_bridge *br, 694 struct br_mrp_start_test *test) 695 { 696 struct br_mrp *mrp = br_mrp_find_id(br, test->ring_id); 697 enum br_mrp_hw_support support; 698 699 if (!mrp) 700 return -EINVAL; 701 702 /* Try to push it to the HW and if it fails then continue with SW 703 * implementation and if that also fails then return error. 704 */ 705 support = br_mrp_switchdev_send_ring_test(br, mrp, test->interval, 706 test->max_miss, test->period, 707 test->monitor); 708 if (support == BR_MRP_NONE) 709 return -EOPNOTSUPP; 710 711 if (support == BR_MRP_HW) 712 return 0; 713 714 mrp->test_interval = test->interval; 715 mrp->test_end = jiffies + usecs_to_jiffies(test->period); 716 mrp->test_max_miss = test->max_miss; 717 mrp->test_monitor = test->monitor; 718 mrp->test_count_miss = 0; 719 queue_delayed_work(system_percpu_wq, &mrp->test_work, 720 usecs_to_jiffies(test->interval)); 721 722 return 0; 723 } 724 725 /* Set in state, int state can be only Open or Closed 726 * note: already called with rtnl_lock 727 */ 728 int br_mrp_set_in_state(struct net_bridge *br, struct br_mrp_in_state *state) 729 { 730 struct br_mrp *mrp = br_mrp_find_in_id(br, state->in_id); 731 732 if (!mrp) 733 return -EINVAL; 734 735 if (mrp->in_state != state->in_state) 736 mrp->in_transitions++; 737 738 mrp->in_state = state->in_state; 739 740 br_mrp_switchdev_set_in_state(br, mrp, state->in_state); 741 742 return 0; 743 } 744 745 /* Set in role, in role can be only MIM(Media Interconnection Manager) or 746 * MIC(Media Interconnection Client). 747 * note: already called with rtnl_lock 748 */ 749 int br_mrp_set_in_role(struct net_bridge *br, struct br_mrp_in_role *role) 750 { 751 struct br_mrp *mrp = br_mrp_find_id(br, role->ring_id); 752 enum br_mrp_hw_support support; 753 struct net_bridge_port *p; 754 755 if (!mrp) 756 return -EINVAL; 757 758 if (!br_mrp_get_port(br, role->i_ifindex)) 759 return -EINVAL; 760 761 if (role->in_role == BR_MRP_IN_ROLE_DISABLED) { 762 u8 state; 763 764 /* It is not allowed to disable a port that doesn't exist */ 765 p = rtnl_dereference(mrp->i_port); 766 if (!p) 767 return -EINVAL; 768 769 /* Stop the generating MRP_InTest frames */ 770 cancel_delayed_work_sync(&mrp->in_test_work); 771 br_mrp_switchdev_send_in_test(br, mrp, 0, 0, 0); 772 773 /* Remove the port */ 774 spin_lock_bh(&br->lock); 775 state = netif_running(br->dev) ? 776 BR_STATE_FORWARDING : BR_STATE_DISABLED; 777 p->state = state; 778 clear_bit(BR_MRP_AWARE_BIT, &p->flags); 779 spin_unlock_bh(&br->lock); 780 br_mrp_port_switchdev_set_state(p, state); 781 rcu_assign_pointer(mrp->i_port, NULL); 782 783 mrp->in_role = role->in_role; 784 mrp->in_id = 0; 785 786 return 0; 787 } 788 789 /* It is not possible to have the same port part of multiple rings */ 790 if (!br_mrp_unique_ifindex(br, role->i_ifindex)) 791 return -EINVAL; 792 793 /* It is not allowed to set a different interconnect port if the mrp 794 * instance has already one. First it needs to be disabled and after 795 * that set the new port 796 */ 797 if (rcu_access_pointer(mrp->i_port)) 798 return -EINVAL; 799 800 p = br_mrp_get_port(br, role->i_ifindex); 801 spin_lock_bh(&br->lock); 802 p->state = BR_STATE_FORWARDING; 803 set_bit(BR_MRP_AWARE_BIT, &p->flags); 804 spin_unlock_bh(&br->lock); 805 rcu_assign_pointer(mrp->i_port, p); 806 807 mrp->in_role = role->in_role; 808 mrp->in_id = role->in_id; 809 810 /* If there is an error just bailed out */ 811 support = br_mrp_switchdev_set_in_role(br, mrp, role->in_id, 812 role->ring_id, role->in_role); 813 if (support == BR_MRP_NONE) 814 return -EOPNOTSUPP; 815 816 /* Now detect if the HW actually applied the role or not. If the HW 817 * applied the role it means that the SW will not to do those operations 818 * anymore. For example if the role is MIM then the HW will notify the 819 * SW when interconnect ring is open, but if the is not pushed to the HW 820 * the SW will need to detect when the interconnect ring is open. 821 */ 822 mrp->in_role_offloaded = support == BR_MRP_SW ? 0 : 1; 823 824 return 0; 825 } 826 827 /* Start to generate MRP_InTest frames, the frames are generated by 828 * HW and if it fails, they are generated by the SW. 829 * note: already called with rtnl_lock 830 */ 831 int br_mrp_start_in_test(struct net_bridge *br, 832 struct br_mrp_start_in_test *in_test) 833 { 834 struct br_mrp *mrp = br_mrp_find_in_id(br, in_test->in_id); 835 enum br_mrp_hw_support support; 836 837 if (!mrp) 838 return -EINVAL; 839 840 if (mrp->in_role != BR_MRP_IN_ROLE_MIM) 841 return -EINVAL; 842 843 /* Try to push it to the HW and if it fails then continue with SW 844 * implementation and if that also fails then return error. 845 */ 846 support = br_mrp_switchdev_send_in_test(br, mrp, in_test->interval, 847 in_test->max_miss, 848 in_test->period); 849 if (support == BR_MRP_NONE) 850 return -EOPNOTSUPP; 851 852 if (support == BR_MRP_HW) 853 return 0; 854 855 mrp->in_test_interval = in_test->interval; 856 mrp->in_test_end = jiffies + usecs_to_jiffies(in_test->period); 857 mrp->in_test_max_miss = in_test->max_miss; 858 mrp->in_test_count_miss = 0; 859 queue_delayed_work(system_percpu_wq, &mrp->in_test_work, 860 usecs_to_jiffies(in_test->interval)); 861 862 return 0; 863 } 864 865 /* Determine if the frame type is a ring frame */ 866 static bool br_mrp_ring_frame(struct sk_buff *skb) 867 { 868 const struct br_mrp_tlv_hdr *hdr; 869 struct br_mrp_tlv_hdr _hdr; 870 871 hdr = skb_header_pointer(skb, sizeof(uint16_t), sizeof(_hdr), &_hdr); 872 if (!hdr) 873 return false; 874 875 if (hdr->type == BR_MRP_TLV_HEADER_RING_TEST || 876 hdr->type == BR_MRP_TLV_HEADER_RING_TOPO || 877 hdr->type == BR_MRP_TLV_HEADER_RING_LINK_DOWN || 878 hdr->type == BR_MRP_TLV_HEADER_RING_LINK_UP || 879 hdr->type == BR_MRP_TLV_HEADER_OPTION) 880 return true; 881 882 return false; 883 } 884 885 /* Determine if the frame type is an interconnect frame */ 886 static bool br_mrp_in_frame(struct sk_buff *skb) 887 { 888 const struct br_mrp_tlv_hdr *hdr; 889 struct br_mrp_tlv_hdr _hdr; 890 891 hdr = skb_header_pointer(skb, sizeof(uint16_t), sizeof(_hdr), &_hdr); 892 if (!hdr) 893 return false; 894 895 if (hdr->type == BR_MRP_TLV_HEADER_IN_TEST || 896 hdr->type == BR_MRP_TLV_HEADER_IN_TOPO || 897 hdr->type == BR_MRP_TLV_HEADER_IN_LINK_DOWN || 898 hdr->type == BR_MRP_TLV_HEADER_IN_LINK_UP || 899 hdr->type == BR_MRP_TLV_HEADER_IN_LINK_STATUS) 900 return true; 901 902 return false; 903 } 904 905 /* Process only MRP Test frame. All the other MRP frames are processed by 906 * userspace application 907 * note: already called with rcu_read_lock 908 */ 909 static void br_mrp_mrm_process(struct br_mrp *mrp, struct net_bridge_port *port, 910 struct sk_buff *skb) 911 { 912 const struct br_mrp_tlv_hdr *hdr; 913 struct br_mrp_tlv_hdr _hdr; 914 915 /* Each MRP header starts with a version field which is 16 bits. 916 * Therefore skip the version and get directly the TLV header. 917 */ 918 hdr = skb_header_pointer(skb, sizeof(uint16_t), sizeof(_hdr), &_hdr); 919 if (!hdr) 920 return; 921 922 if (hdr->type != BR_MRP_TLV_HEADER_RING_TEST) 923 return; 924 925 mrp->test_count_miss = 0; 926 927 /* Notify the userspace that the ring is closed only when the ring is 928 * not closed 929 */ 930 if (mrp->ring_state != BR_MRP_RING_STATE_CLOSED) 931 br_mrp_ring_port_open(port->dev, false); 932 } 933 934 /* Determine if the test hdr has a better priority than the node */ 935 static bool br_mrp_test_better_than_own(struct br_mrp *mrp, 936 struct net_bridge *br, 937 const struct br_mrp_ring_test_hdr *hdr) 938 { 939 u16 prio = be16_to_cpu(hdr->prio); 940 941 if (prio < mrp->prio || 942 (prio == mrp->prio && 943 ether_addr_to_u64(hdr->sa) < ether_addr_to_u64(br->dev->dev_addr))) 944 return true; 945 946 return false; 947 } 948 949 /* Process only MRP Test frame. All the other MRP frames are processed by 950 * userspace application 951 * note: already called with rcu_read_lock 952 */ 953 static void br_mrp_mra_process(struct br_mrp *mrp, struct net_bridge *br, 954 struct net_bridge_port *port, 955 struct sk_buff *skb) 956 { 957 const struct br_mrp_ring_test_hdr *test_hdr; 958 struct br_mrp_ring_test_hdr _test_hdr; 959 const struct br_mrp_tlv_hdr *hdr; 960 struct br_mrp_tlv_hdr _hdr; 961 962 /* Each MRP header starts with a version field which is 16 bits. 963 * Therefore skip the version and get directly the TLV header. 964 */ 965 hdr = skb_header_pointer(skb, sizeof(uint16_t), sizeof(_hdr), &_hdr); 966 if (!hdr) 967 return; 968 969 if (hdr->type != BR_MRP_TLV_HEADER_RING_TEST) 970 return; 971 972 test_hdr = skb_header_pointer(skb, sizeof(uint16_t) + sizeof(_hdr), 973 sizeof(_test_hdr), &_test_hdr); 974 if (!test_hdr) 975 return; 976 977 /* Only frames that have a better priority than the node will 978 * clear the miss counter because otherwise the node will need to behave 979 * as MRM. 980 */ 981 if (br_mrp_test_better_than_own(mrp, br, test_hdr)) 982 mrp->test_count_miss = 0; 983 } 984 985 /* Process only MRP InTest frame. All the other MRP frames are processed by 986 * userspace application 987 * note: already called with rcu_read_lock 988 */ 989 static bool br_mrp_mim_process(struct br_mrp *mrp, struct net_bridge_port *port, 990 struct sk_buff *skb) 991 { 992 const struct br_mrp_in_test_hdr *in_hdr; 993 struct br_mrp_in_test_hdr _in_hdr; 994 const struct br_mrp_tlv_hdr *hdr; 995 struct br_mrp_tlv_hdr _hdr; 996 997 /* Each MRP header starts with a version field which is 16 bits. 998 * Therefore skip the version and get directly the TLV header. 999 */ 1000 hdr = skb_header_pointer(skb, sizeof(uint16_t), sizeof(_hdr), &_hdr); 1001 if (!hdr) 1002 return false; 1003 1004 /* The check for InTest frame type was already done */ 1005 in_hdr = skb_header_pointer(skb, sizeof(uint16_t) + sizeof(_hdr), 1006 sizeof(_in_hdr), &_in_hdr); 1007 if (!in_hdr) 1008 return false; 1009 1010 /* It needs to process only it's own InTest frames. */ 1011 if (mrp->in_id != ntohs(in_hdr->id)) 1012 return false; 1013 1014 mrp->in_test_count_miss = 0; 1015 1016 /* Notify the userspace that the ring is closed only when the ring is 1017 * not closed 1018 */ 1019 if (mrp->in_state != BR_MRP_IN_STATE_CLOSED) 1020 br_mrp_in_port_open(port->dev, false); 1021 1022 return true; 1023 } 1024 1025 /* Get the MRP frame type 1026 * note: already called with rcu_read_lock 1027 */ 1028 static u8 br_mrp_get_frame_type(struct sk_buff *skb) 1029 { 1030 const struct br_mrp_tlv_hdr *hdr; 1031 struct br_mrp_tlv_hdr _hdr; 1032 1033 /* Each MRP header starts with a version field which is 16 bits. 1034 * Therefore skip the version and get directly the TLV header. 1035 */ 1036 hdr = skb_header_pointer(skb, sizeof(uint16_t), sizeof(_hdr), &_hdr); 1037 if (!hdr) 1038 return 0xff; 1039 1040 return hdr->type; 1041 } 1042 1043 static bool br_mrp_mrm_behaviour(struct br_mrp *mrp) 1044 { 1045 if (mrp->ring_role == BR_MRP_RING_ROLE_MRM || 1046 (mrp->ring_role == BR_MRP_RING_ROLE_MRA && !mrp->test_monitor)) 1047 return true; 1048 1049 return false; 1050 } 1051 1052 static bool br_mrp_mrc_behaviour(struct br_mrp *mrp) 1053 { 1054 if (mrp->ring_role == BR_MRP_RING_ROLE_MRC || 1055 (mrp->ring_role == BR_MRP_RING_ROLE_MRA && mrp->test_monitor)) 1056 return true; 1057 1058 return false; 1059 } 1060 1061 /* This will just forward the frame to the other mrp ring ports, depending on 1062 * the frame type, ring role and interconnect role 1063 * note: already called with rcu_read_lock 1064 */ 1065 static int br_mrp_rcv(struct net_bridge_port *p, 1066 struct sk_buff *skb, struct net_device *dev) 1067 { 1068 struct net_bridge_port *p_port, *s_port, *i_port = NULL; 1069 struct net_bridge_port *p_dst, *s_dst, *i_dst = NULL; 1070 struct net_bridge *br; 1071 struct br_mrp *mrp; 1072 1073 /* If port is disabled don't accept any frames */ 1074 if (p->state == BR_STATE_DISABLED) 1075 return 0; 1076 1077 br = p->br; 1078 mrp = br_mrp_find_port(br, p); 1079 if (unlikely(!mrp)) 1080 return 0; 1081 1082 p_port = rcu_dereference(mrp->p_port); 1083 if (!p_port) 1084 return 0; 1085 p_dst = p_port; 1086 1087 s_port = rcu_dereference(mrp->s_port); 1088 if (!s_port) 1089 return 0; 1090 s_dst = s_port; 1091 1092 /* If the frame is a ring frame then it is not required to check the 1093 * interconnect role and ports to process or forward the frame 1094 */ 1095 if (br_mrp_ring_frame(skb)) { 1096 /* If the role is MRM then don't forward the frames */ 1097 if (mrp->ring_role == BR_MRP_RING_ROLE_MRM) { 1098 br_mrp_mrm_process(mrp, p, skb); 1099 goto no_forward; 1100 } 1101 1102 /* If the role is MRA then don't forward the frames if it 1103 * behaves as MRM node 1104 */ 1105 if (mrp->ring_role == BR_MRP_RING_ROLE_MRA) { 1106 if (!mrp->test_monitor) { 1107 br_mrp_mrm_process(mrp, p, skb); 1108 goto no_forward; 1109 } 1110 1111 br_mrp_mra_process(mrp, br, p, skb); 1112 } 1113 1114 goto forward; 1115 } 1116 1117 if (br_mrp_in_frame(skb)) { 1118 u8 in_type = br_mrp_get_frame_type(skb); 1119 1120 i_port = rcu_dereference(mrp->i_port); 1121 i_dst = i_port; 1122 1123 /* If the ring port is in block state it should not forward 1124 * In_Test frames 1125 */ 1126 if (br_mrp_is_ring_port(p_port, s_port, p) && 1127 p->state == BR_STATE_BLOCKING && 1128 in_type == BR_MRP_TLV_HEADER_IN_TEST) 1129 goto no_forward; 1130 1131 /* Nodes that behaves as MRM needs to stop forwarding the 1132 * frames in case the ring is closed, otherwise will be a loop. 1133 * In this case the frame is no forward between the ring ports. 1134 */ 1135 if (br_mrp_mrm_behaviour(mrp) && 1136 br_mrp_is_ring_port(p_port, s_port, p) && 1137 (s_port->state != BR_STATE_FORWARDING || 1138 p_port->state != BR_STATE_FORWARDING)) { 1139 p_dst = NULL; 1140 s_dst = NULL; 1141 } 1142 1143 /* A node that behaves as MRC and doesn't have a interconnect 1144 * role then it should forward all frames between the ring ports 1145 * because it doesn't have an interconnect port 1146 */ 1147 if (br_mrp_mrc_behaviour(mrp) && 1148 mrp->in_role == BR_MRP_IN_ROLE_DISABLED) 1149 goto forward; 1150 1151 if (mrp->in_role == BR_MRP_IN_ROLE_MIM) { 1152 if (in_type == BR_MRP_TLV_HEADER_IN_TEST) { 1153 /* MIM should not forward it's own InTest 1154 * frames 1155 */ 1156 if (br_mrp_mim_process(mrp, p, skb)) { 1157 goto no_forward; 1158 } else { 1159 if (br_mrp_is_ring_port(p_port, s_port, 1160 p)) 1161 i_dst = NULL; 1162 1163 if (br_mrp_is_in_port(i_port, p)) 1164 goto no_forward; 1165 } 1166 } else { 1167 /* MIM should forward IntLinkChange/Status and 1168 * IntTopoChange between ring ports but MIM 1169 * should not forward IntLinkChange/Status and 1170 * IntTopoChange if the frame was received at 1171 * the interconnect port 1172 */ 1173 if (br_mrp_is_ring_port(p_port, s_port, p)) 1174 i_dst = NULL; 1175 1176 if (br_mrp_is_in_port(i_port, p)) 1177 goto no_forward; 1178 } 1179 } 1180 1181 if (mrp->in_role == BR_MRP_IN_ROLE_MIC) { 1182 /* MIC should forward InTest frames on all ports 1183 * regardless of the received port 1184 */ 1185 if (in_type == BR_MRP_TLV_HEADER_IN_TEST) 1186 goto forward; 1187 1188 /* MIC should forward IntLinkChange frames only if they 1189 * are received on ring ports to all the ports 1190 */ 1191 if (br_mrp_is_ring_port(p_port, s_port, p) && 1192 (in_type == BR_MRP_TLV_HEADER_IN_LINK_UP || 1193 in_type == BR_MRP_TLV_HEADER_IN_LINK_DOWN)) 1194 goto forward; 1195 1196 /* MIC should forward IntLinkStatus frames only to 1197 * interconnect port if it was received on a ring port. 1198 * If it is received on interconnect port then, it 1199 * should be forward on both ring ports 1200 */ 1201 if (br_mrp_is_ring_port(p_port, s_port, p) && 1202 in_type == BR_MRP_TLV_HEADER_IN_LINK_STATUS) { 1203 p_dst = NULL; 1204 s_dst = NULL; 1205 } 1206 1207 /* Should forward the InTopo frames only between the 1208 * ring ports 1209 */ 1210 if (in_type == BR_MRP_TLV_HEADER_IN_TOPO) { 1211 i_dst = NULL; 1212 goto forward; 1213 } 1214 1215 /* In all the other cases don't forward the frames */ 1216 goto no_forward; 1217 } 1218 } 1219 1220 forward: 1221 if (p_dst) 1222 br_forward(p_dst, skb, true, false); 1223 if (s_dst) 1224 br_forward(s_dst, skb, true, false); 1225 if (i_dst) 1226 br_forward(i_dst, skb, true, false); 1227 1228 no_forward: 1229 return 1; 1230 } 1231 1232 /* Check if the frame was received on a port that is part of MRP ring 1233 * and if the frame has MRP eth. In that case process the frame otherwise do 1234 * normal forwarding. 1235 * note: already called with rcu_read_lock 1236 */ 1237 int br_mrp_process(struct net_bridge_port *p, struct sk_buff *skb) 1238 { 1239 /* If there is no MRP instance do normal forwarding */ 1240 if (likely(!test_bit(BR_MRP_AWARE_BIT, &p->flags))) 1241 goto out; 1242 1243 return br_mrp_rcv(p, skb, p->dev); 1244 out: 1245 return 0; 1246 } 1247 1248 bool br_mrp_enabled(struct net_bridge *br) 1249 { 1250 return !hlist_empty(&br->mrp_list); 1251 } 1252