FILE
nexthop.c
File preview with syntax highlighting.
linux-6.19.14/net/ipv4/nexthop.c
1// SPDX-License-Identifier: GPL-2.0 2/* Generic nexthop implementation 3 * 4 * Copyright (c) 2017-19 Cumulus Networks 5 * Copyright (c) 2017-19 David Ahern <dsa@cumulusnetworks.com> 6 */ 7 8#include <linux/nexthop.h> 9#include <linux/rtnetlink.h> 10#include <linux/slab.h> 11#include <linux/vmalloc.h> 12#include <net/arp.h> 13#include <net/ipv6_stubs.h> 14#include <net/lwtunnel.h> 15#include <net/ndisc.h> 16#include <net/nexthop.h> 17#include <net/route.h> 18#include <net/sock.h> 19 20#define NH_RES_DEFAULT_IDLE_TIMER (120 * HZ) 21#define NH_RES_DEFAULT_UNBALANCED_TIMER 0 /* No forced rebalancing. */ 22 23static void remove_nexthop(struct net *net, struct nexthop *nh, 24 struct nl_info *nlinfo); 25 26#define NH_DEV_HASHBITS 8 27#define NH_DEV_HASHSIZE (1U << NH_DEV_HASHBITS) 28 29#define NHA_OP_FLAGS_DUMP_ALL (NHA_OP_FLAG_DUMP_STATS | \ 30 NHA_OP_FLAG_DUMP_HW_STATS) 31 32static const struct nla_policy rtm_nh_policy_new[] = { 33 [NHA_ID] = { .type = NLA_U32 }, 34 [NHA_GROUP] = { .type = NLA_BINARY }, 35 [NHA_GROUP_TYPE] = { .type = NLA_U16 }, 36 [NHA_BLACKHOLE] = { .type = NLA_FLAG }, 37 [NHA_OIF] = { .type = NLA_U32 }, 38 [NHA_GATEWAY] = { .type = NLA_BINARY }, 39 [NHA_ENCAP_TYPE] = { .type = NLA_U16 }, 40 [NHA_ENCAP] = { .type = NLA_NESTED }, 41 [NHA_FDB] = { .type = NLA_FLAG }, 42 [NHA_RES_GROUP] = { .type = NLA_NESTED }, 43 [NHA_HW_STATS_ENABLE] = NLA_POLICY_MAX(NLA_U32, true), 44}; 45 46static const struct nla_policy rtm_nh_policy_get[] = { 47 [NHA_ID] = { .type = NLA_U32 }, 48 [NHA_OP_FLAGS] = NLA_POLICY_MASK(NLA_U32, 49 NHA_OP_FLAGS_DUMP_ALL), 50}; 51 52static const struct nla_policy rtm_nh_policy_del[] = { 53 [NHA_ID] = { .type = NLA_U32 }, 54}; 55 56static const struct nla_policy rtm_nh_policy_dump[] = { 57 [NHA_OIF] = { .type = NLA_U32 }, 58 [NHA_GROUPS] = { .type = NLA_FLAG }, 59 [NHA_MASTER] = { .type = NLA_U32 }, 60 [NHA_FDB] = { .type = NLA_FLAG }, 61 [NHA_OP_FLAGS] = NLA_POLICY_MASK(NLA_U32, 62 NHA_OP_FLAGS_DUMP_ALL), 63}; 64 65static const struct nla_policy rtm_nh_res_policy_new[] = { 66 [NHA_RES_GROUP_BUCKETS] = { .type = NLA_U16 }, 67 [NHA_RES_GROUP_IDLE_TIMER] = { .type = NLA_U32 }, 68 [NHA_RES_GROUP_UNBALANCED_TIMER] = { .type = NLA_U32 }, 69}; 70 71static const struct nla_policy rtm_nh_policy_dump_bucket[] = { 72 [NHA_ID] = { .type = NLA_U32 }, 73 [NHA_OIF] = { .type = NLA_U32 }, 74 [NHA_MASTER] = { .type = NLA_U32 }, 75 [NHA_RES_BUCKET] = { .type = NLA_NESTED }, 76}; 77 78static const struct nla_policy rtm_nh_res_bucket_policy_dump[] = { 79 [NHA_RES_BUCKET_NH_ID] = { .type = NLA_U32 }, 80}; 81 82static const struct nla_policy rtm_nh_policy_get_bucket[] = { 83 [NHA_ID] = { .type = NLA_U32 }, 84 [NHA_RES_BUCKET] = { .type = NLA_NESTED }, 85}; 86 87static const struct nla_policy rtm_nh_res_bucket_policy_get[] = { 88 [NHA_RES_BUCKET_INDEX] = { .type = NLA_U16 }, 89}; 90 91static bool nexthop_notifiers_is_empty(struct net *net) 92{ 93 return !net->nexthop.notifier_chain.head; 94} 95 96static void 97__nh_notifier_single_info_init(struct nh_notifier_single_info *nh_info, 98 const struct nh_info *nhi) 99{ 100 nh_info->dev = nhi->fib_nhc.nhc_dev; 101 nh_info->gw_family = nhi->fib_nhc.nhc_gw_family; 102 if (nh_info->gw_family == AF_INET) 103 nh_info->ipv4 = nhi->fib_nhc.nhc_gw.ipv4; 104 else if (nh_info->gw_family == AF_INET6) 105 nh_info->ipv6 = nhi->fib_nhc.nhc_gw.ipv6; 106 107 nh_info->id = nhi->nh_parent->id; 108 nh_info->is_reject = nhi->reject_nh; 109 nh_info->is_fdb = nhi->fdb_nh; 110 nh_info->has_encap = !!nhi->fib_nhc.nhc_lwtstate; 111} 112 113static int nh_notifier_single_info_init(struct nh_notifier_info *info, 114 const struct nexthop *nh) 115{ 116 struct nh_info *nhi = rtnl_dereference(nh->nh_info); 117 118 info->type = NH_NOTIFIER_INFO_TYPE_SINGLE; 119 info->nh = kzalloc(sizeof(*info->nh), GFP_KERNEL); 120 if (!info->nh) 121 return -ENOMEM; 122 123 __nh_notifier_single_info_init(info->nh, nhi); 124 125 return 0; 126} 127 128static void nh_notifier_single_info_fini(struct nh_notifier_info *info) 129{ 130 kfree(info->nh); 131} 132 133static int nh_notifier_mpath_info_init(struct nh_notifier_info *info, 134 struct nh_group *nhg) 135{ 136 u16 num_nh = nhg->num_nh; 137 int i; 138 139 info->type = NH_NOTIFIER_INFO_TYPE_GRP; 140 info->nh_grp = kzalloc(struct_size(info->nh_grp, nh_entries, num_nh), 141 GFP_KERNEL); 142 if (!info->nh_grp) 143 return -ENOMEM; 144 145 info->nh_grp->num_nh = num_nh; 146 info->nh_grp->is_fdb = nhg->fdb_nh; 147 info->nh_grp->hw_stats = nhg->hw_stats; 148 149 for (i = 0; i < num_nh; i++) { 150 struct nh_grp_entry *nhge = &nhg->nh_entries[i]; 151 struct nh_info *nhi; 152 153 nhi = rtnl_dereference(nhge->nh->nh_info); 154 info->nh_grp->nh_entries[i].weight = nhge->weight; 155 __nh_notifier_single_info_init(&info->nh_grp->nh_entries[i].nh, 156 nhi); 157 } 158 159 return 0; 160} 161 162static int nh_notifier_res_table_info_init(struct nh_notifier_info *info, 163 struct nh_group *nhg) 164{ 165 struct nh_res_table *res_table = rtnl_dereference(nhg->res_table); 166 u16 num_nh_buckets = res_table->num_nh_buckets; 167 unsigned long size; 168 u16 i; 169 170 info->type = NH_NOTIFIER_INFO_TYPE_RES_TABLE; 171 size = struct_size(info->nh_res_table, nhs, num_nh_buckets); 172 info->nh_res_table = __vmalloc(size, GFP_KERNEL | __GFP_ZERO | 173 __GFP_NOWARN); 174 if (!info->nh_res_table) 175 return -ENOMEM; 176 177 info->nh_res_table->num_nh_buckets = num_nh_buckets; 178 info->nh_res_table->hw_stats = nhg->hw_stats; 179 180 for (i = 0; i < num_nh_buckets; i++) { 181 struct nh_res_bucket *bucket = &res_table->nh_buckets[i]; 182 struct nh_grp_entry *nhge; 183 struct nh_info *nhi; 184 185 nhge = rtnl_dereference(bucket->nh_entry); 186 nhi = rtnl_dereference(nhge->nh->nh_info); 187 __nh_notifier_single_info_init(&info->nh_res_table->nhs[i], 188 nhi); 189 } 190 191 return 0; 192} 193 194static int nh_notifier_grp_info_init(struct nh_notifier_info *info, 195 const struct nexthop *nh) 196{ 197 struct nh_group *nhg = rtnl_dereference(nh->nh_grp); 198 199 if (nhg->hash_threshold) 200 return nh_notifier_mpath_info_init(info, nhg); 201 else if (nhg->resilient) 202 return nh_notifier_res_table_info_init(info, nhg); 203 return -EINVAL; 204} 205 206static void nh_notifier_grp_info_fini(struct nh_notifier_info *info, 207 const struct nexthop *nh) 208{ 209 struct nh_group *nhg = rtnl_dereference(nh->nh_grp); 210 211 if (nhg->hash_threshold) 212 kfree(info->nh_grp); 213 else if (nhg->resilient) 214 vfree(info->nh_res_table); 215} 216 217static int nh_notifier_info_init(struct nh_notifier_info *info, 218 const struct nexthop *nh) 219{ 220 info->id = nh->id; 221 222 if (nh->is_group) 223 return nh_notifier_grp_info_init(info, nh); 224 else 225 return nh_notifier_single_info_init(info, nh); 226} 227 228static void nh_notifier_info_fini(struct nh_notifier_info *info, 229 const struct nexthop *nh) 230{ 231 if (nh->is_group) 232 nh_notifier_grp_info_fini(info, nh); 233 else 234 nh_notifier_single_info_fini(info); 235} 236 237static int call_nexthop_notifiers(struct net *net, 238 enum nexthop_event_type event_type, 239 struct nexthop *nh, 240 struct netlink_ext_ack *extack) 241{ 242 struct nh_notifier_info info = { 243 .net = net, 244 .extack = extack, 245 }; 246 int err; 247 248 ASSERT_RTNL(); 249 250 if (nexthop_notifiers_is_empty(net)) 251 return 0; 252 253 err = nh_notifier_info_init(&info, nh); 254 if (err) { 255 NL_SET_ERR_MSG(extack, "Failed to initialize nexthop notifier info"); 256 return err; 257 } 258 259 err = blocking_notifier_call_chain(&net->nexthop.notifier_chain, 260 event_type, &info); 261 nh_notifier_info_fini(&info, nh); 262 263 return notifier_to_errno(err); 264} 265 266static int 267nh_notifier_res_bucket_idle_timer_get(const struct nh_notifier_info *info, 268 bool force, unsigned int *p_idle_timer_ms) 269{ 270 struct nh_res_table *res_table; 271 struct nh_group *nhg; 272 struct nexthop *nh; 273 int err = 0; 274 275 /* When 'force' is false, nexthop bucket replacement is performed 276 * because the bucket was deemed to be idle. In this case, capable 277 * listeners can choose to perform an atomic replacement: The bucket is 278 * only replaced if it is inactive. However, if the idle timer interval 279 * is smaller than the interval in which a listener is querying 280 * buckets' activity from the device, then atomic replacement should 281 * not be tried. Pass the idle timer value to listeners, so that they 282 * could determine which type of replacement to perform. 283 */ 284 if (force) { 285 *p_idle_timer_ms = 0; 286 return 0; 287 } 288 289 rcu_read_lock(); 290 291 nh = nexthop_find_by_id(info->net, info->id); 292 if (!nh) { 293 err = -EINVAL; 294 goto out; 295 } 296 297 nhg = rcu_dereference(nh->nh_grp); 298 res_table = rcu_dereference(nhg->res_table); 299 *p_idle_timer_ms = jiffies_to_msecs(res_table->idle_timer); 300 301out: 302 rcu_read_unlock(); 303 304 return err; 305} 306 307static int nh_notifier_res_bucket_info_init(struct nh_notifier_info *info, 308 u16 bucket_index, bool force, 309 struct nh_info *oldi, 310 struct nh_info *newi) 311{ 312 unsigned int idle_timer_ms; 313 int err; 314 315 err = nh_notifier_res_bucket_idle_timer_get(info, force, 316 &idle_timer_ms); 317 if (err) 318 return err; 319 320 info->type = NH_NOTIFIER_INFO_TYPE_RES_BUCKET; 321 info->nh_res_bucket = kzalloc(sizeof(*info->nh_res_bucket), 322 GFP_KERNEL); 323 if (!info->nh_res_bucket) 324 return -ENOMEM; 325 326 info->nh_res_bucket->bucket_index = bucket_index; 327 info->nh_res_bucket->idle_timer_ms = idle_timer_ms; 328 info->nh_res_bucket->force = force; 329 __nh_notifier_single_info_init(&info->nh_res_bucket->old_nh, oldi); 330 __nh_notifier_single_info_init(&info->nh_res_bucket->new_nh, newi); 331 return 0; 332} 333 334static void nh_notifier_res_bucket_info_fini(struct nh_notifier_info *info) 335{ 336 kfree(info->nh_res_bucket); 337} 338 339static int __call_nexthop_res_bucket_notifiers(struct net *net, u32 nhg_id, 340 u16 bucket_index, bool force, 341 struct nh_info *oldi, 342 struct nh_info *newi, 343 struct netlink_ext_ack *extack) 344{ 345 struct nh_notifier_info info = { 346 .net = net, 347 .extack = extack, 348 .id = nhg_id, 349 }; 350 int err; 351 352 if (nexthop_notifiers_is_empty(net)) 353 return 0; 354 355 err = nh_notifier_res_bucket_info_init(&info, bucket_index, force, 356 oldi, newi); 357 if (err) 358 return err; 359 360 err = blocking_notifier_call_chain(&net->nexthop.notifier_chain, 361 NEXTHOP_EVENT_BUCKET_REPLACE, &info); 362 nh_notifier_res_bucket_info_fini(&info); 363 364 return notifier_to_errno(err); 365} 366 367/* There are three users of RES_TABLE, and NHs etc. referenced from there: 368 * 369 * 1) a collection of callbacks for NH maintenance. This operates under 370 * RTNL, 371 * 2) the delayed work that gradually balances the resilient table, 372 * 3) and nexthop_select_path(), operating under RCU. 373 * 374 * Both the delayed work and the RTNL block are writers, and need to 375 * maintain mutual exclusion. Since there are only two and well-known 376 * writers for each table, the RTNL code can make sure it has exclusive 377 * access thus: 378 * 379 * - Have the DW operate without locking; 380 * - synchronously cancel the DW; 381 * - do the writing; 382 * - if the write was not actually a delete, call upkeep, which schedules 383 * DW again if necessary. 384 * 385 * The functions that are always called from the RTNL context use 386 * rtnl_dereference(). The functions that can also be called from the DW do 387 * a raw dereference and rely on the above mutual exclusion scheme. 388 */ 389#define nh_res_dereference(p) (rcu_dereference_raw(p)) 390 391static int call_nexthop_res_bucket_notifiers(struct net *net, u32 nhg_id, 392 u16 bucket_index, bool force, 393 struct nexthop *old_nh, 394 struct nexthop *new_nh, 395 struct netlink_ext_ack *extack) 396{ 397 struct nh_info *oldi = nh_res_dereference(old_nh->nh_info); 398 struct nh_info *newi = nh_res_dereference(new_nh->nh_info); 399 400 return __call_nexthop_res_bucket_notifiers(net, nhg_id, bucket_index, 401 force, oldi, newi, extack); 402} 403 404static int call_nexthop_res_table_notifiers(struct net *net, struct nexthop *nh, 405 struct netlink_ext_ack *extack) 406{ 407 struct nh_notifier_info info = { 408 .net = net, 409 .extack = extack, 410 .id = nh->id, 411 }; 412 struct nh_group *nhg; 413 int err; 414 415 ASSERT_RTNL(); 416 417 if (nexthop_notifiers_is_empty(net)) 418 return 0; 419 420 /* At this point, the nexthop buckets are still not populated. Only 421 * emit a notification with the logical nexthops, so that a listener 422 * could potentially veto it in case of unsupported configuration. 423 */ 424 nhg = rtnl_dereference(nh->nh_grp); 425 err = nh_notifier_mpath_info_init(&info, nhg); 426 if (err) { 427 NL_SET_ERR_MSG(extack, "Failed to initialize nexthop notifier info"); 428 return err; 429 } 430 431 err = blocking_notifier_call_chain(&net->nexthop.notifier_chain, 432 NEXTHOP_EVENT_RES_TABLE_PRE_REPLACE, 433 &info); 434 kfree(info.nh_grp); 435 436 return notifier_to_errno(err); 437} 438 439static int call_nexthop_notifier(struct notifier_block *nb, struct net *net, 440 enum nexthop_event_type event_type, 441 struct nexthop *nh, 442 struct netlink_ext_ack *extack) 443{ 444 struct nh_notifier_info info = { 445 .net = net, 446 .extack = extack, 447 }; 448 int err; 449 450 err = nh_notifier_info_init(&info, nh); 451 if (err) 452 return err; 453 454 err = nb->notifier_call(nb, event_type, &info); 455 nh_notifier_info_fini(&info, nh); 456 457 return notifier_to_errno(err); 458} 459 460static unsigned int nh_dev_hashfn(unsigned int val) 461{ 462 unsigned int mask = NH_DEV_HASHSIZE - 1; 463 464 return (val ^ 465 (val >> NH_DEV_HASHBITS) ^ 466 (val >> (NH_DEV_HASHBITS * 2))) & mask; 467} 468 469static void nexthop_devhash_add(struct net *net, struct nh_info *nhi) 470{ 471 struct net_device *dev = nhi->fib_nhc.nhc_dev; 472 struct hlist_head *head; 473 unsigned int hash; 474 475 WARN_ON(!dev); 476 477 hash = nh_dev_hashfn(dev->ifindex); 478 head = &net->nexthop.devhash[hash]; 479 hlist_add_head(&nhi->dev_hash, head); 480} 481 482static void nexthop_free_group(struct nexthop *nh) 483{ 484 struct nh_group *nhg; 485 int i; 486 487 nhg = rcu_dereference_raw(nh->nh_grp); 488 for (i = 0; i < nhg->num_nh; ++i) { 489 struct nh_grp_entry *nhge = &nhg->nh_entries[i]; 490 491 WARN_ON(!list_empty(&nhge->nh_list)); 492 free_percpu(nhge->stats); 493 nexthop_put(nhge->nh); 494 } 495 496 WARN_ON(nhg->spare == nhg); 497 498 if (nhg->resilient) 499 vfree(rcu_dereference_raw(nhg->res_table)); 500 501 kfree(nhg->spare); 502 kfree(nhg); 503} 504 505static void nexthop_free_single(struct nexthop *nh) 506{ 507 struct nh_info *nhi; 508 509 nhi = rcu_dereference_raw(nh->nh_info); 510 switch (nhi->family) { 511 case AF_INET: 512 fib_nh_release(nh->net, &nhi->fib_nh); 513 break; 514 case AF_INET6: 515 ipv6_stub->fib6_nh_release(&nhi->fib6_nh); 516 break; 517 } 518 kfree(nhi); 519} 520 521void nexthop_free_rcu(struct rcu_head *head) 522{ 523 struct nexthop *nh = container_of(head, struct nexthop, rcu); 524 525 if (nh->is_group) 526 nexthop_free_group(nh); 527 else 528 nexthop_free_single(nh); 529 530 kfree(nh); 531} 532EXPORT_SYMBOL_GPL(nexthop_free_rcu); 533 534static struct nexthop *nexthop_alloc(void) 535{ 536 struct nexthop *nh; 537 538 nh = kzalloc(sizeof(struct nexthop), GFP_KERNEL); 539 if (nh) { 540 INIT_LIST_HEAD(&nh->fi_list); 541 INIT_LIST_HEAD(&nh->f6i_list); 542 INIT_LIST_HEAD(&nh->grp_list); 543 INIT_LIST_HEAD(&nh->fdb_list); 544 spin_lock_init(&nh->lock); 545 } 546 return nh; 547} 548 549static struct nh_group *nexthop_grp_alloc(u16 num_nh) 550{ 551 struct nh_group *nhg; 552 553 nhg = kzalloc(struct_size(nhg, nh_entries, num_nh), GFP_KERNEL); 554 if (nhg) 555 nhg->num_nh = num_nh; 556 557 return nhg; 558} 559 560static void nh_res_table_upkeep_dw(struct work_struct *work); 561 562static struct nh_res_table * 563nexthop_res_table_alloc(struct net *net, u32 nhg_id, struct nh_config *cfg) 564{ 565 const u16 num_nh_buckets = cfg->nh_grp_res_num_buckets; 566 struct nh_res_table *res_table; 567 unsigned long size; 568 569 size = struct_size(res_table, nh_buckets, num_nh_buckets); 570 res_table = __vmalloc(size, GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN); 571 if (!res_table) 572 return NULL; 573 574 res_table->net = net; 575 res_table->nhg_id = nhg_id; 576 INIT_DELAYED_WORK(&res_table->upkeep_dw, &nh_res_table_upkeep_dw); 577 INIT_LIST_HEAD(&res_table->uw_nh_entries); 578 res_table->idle_timer = cfg->nh_grp_res_idle_timer; 579 res_table->unbalanced_timer = cfg->nh_grp_res_unbalanced_timer; 580 res_table->num_nh_buckets = num_nh_buckets; 581 return res_table; 582} 583 584static void nh_base_seq_inc(struct net *net) 585{ 586 while (++net->nexthop.seq == 0) 587 ; 588} 589 590/* no reference taken; rcu lock or rtnl must be held */ 591struct nexthop *nexthop_find_by_id(struct net *net, u32 id) 592{ 593 struct rb_node **pp, *parent = NULL, *next; 594 595 pp = &net->nexthop.rb_root.rb_node; 596 while (1) { 597 struct nexthop *nh; 598 599 next = rcu_dereference_raw(*pp); 600 if (!next) 601 break; 602 parent = next; 603 604 nh = rb_entry(parent, struct nexthop, rb_node); 605 if (id < nh->id) 606 pp = &next->rb_left; 607 else if (id > nh->id) 608 pp = &next->rb_right; 609 else 610 return nh; 611 } 612 return NULL; 613} 614EXPORT_SYMBOL_GPL(nexthop_find_by_id); 615 616/* used for auto id allocation; called with rtnl held */ 617static u32 nh_find_unused_id(struct net *net) 618{ 619 u32 id_start = net->nexthop.last_id_allocated; 620 621 while (1) { 622 net->nexthop.last_id_allocated++; 623 if (net->nexthop.last_id_allocated == id_start) 624 break; 625 626 if (!nexthop_find_by_id(net, net->nexthop.last_id_allocated)) 627 return net->nexthop.last_id_allocated; 628 } 629 return 0; 630} 631 632static void nh_res_time_set_deadline(unsigned long next_time, 633 unsigned long *deadline) 634{ 635 if (time_before(next_time, *deadline)) 636 *deadline = next_time; 637} 638 639static clock_t nh_res_table_unbalanced_time(struct nh_res_table *res_table) 640{ 641 if (list_empty(&res_table->uw_nh_entries)) 642 return 0; 643 return jiffies_delta_to_clock_t(jiffies - res_table->unbalanced_since); 644} 645 646static int nla_put_nh_group_res(struct sk_buff *skb, struct nh_group *nhg) 647{ 648 struct nh_res_table *res_table = rtnl_dereference(nhg->res_table); 649 struct nlattr *nest; 650 651 nest = nla_nest_start(skb, NHA_RES_GROUP); 652 if (!nest) 653 return -EMSGSIZE; 654 655 if (nla_put_u16(skb, NHA_RES_GROUP_BUCKETS, 656 res_table->num_nh_buckets) || 657 nla_put_u32(skb, NHA_RES_GROUP_IDLE_TIMER, 658 jiffies_to_clock_t(res_table->idle_timer)) || 659 nla_put_u32(skb, NHA_RES_GROUP_UNBALANCED_TIMER, 660 jiffies_to_clock_t(res_table->unbalanced_timer)) || 661 nla_put_u64_64bit(skb, NHA_RES_GROUP_UNBALANCED_TIME, 662 nh_res_table_unbalanced_time(res_table), 663 NHA_RES_GROUP_PAD)) 664 goto nla_put_failure; 665 666 nla_nest_end(skb, nest); 667 return 0; 668 669nla_put_failure: 670 nla_nest_cancel(skb, nest); 671 return -EMSGSIZE; 672} 673 674static void nh_grp_entry_stats_inc(struct nh_grp_entry *nhge) 675{ 676 struct nh_grp_entry_stats *cpu_stats; 677 678 cpu_stats = get_cpu_ptr(nhge->stats); 679 u64_stats_update_begin(&cpu_stats->syncp); 680 u64_stats_inc(&cpu_stats->packets); 681 u64_stats_update_end(&cpu_stats->syncp); 682 put_cpu_ptr(cpu_stats); 683} 684 685static void nh_grp_entry_stats_read(struct nh_grp_entry *nhge, 686 u64 *ret_packets) 687{ 688 int i; 689 690 *ret_packets = 0; 691 692 for_each_possible_cpu(i) { 693 struct nh_grp_entry_stats *cpu_stats; 694 unsigned int start; 695 u64 packets; 696 697 cpu_stats = per_cpu_ptr(nhge->stats, i); 698 do { 699 start = u64_stats_fetch_begin(&cpu_stats->syncp); 700 packets = u64_stats_read(&cpu_stats->packets); 701 } while (u64_stats_fetch_retry(&cpu_stats->syncp, start)); 702 703 *ret_packets += packets; 704 } 705} 706 707static int nh_notifier_grp_hw_stats_init(struct nh_notifier_info *info, 708 const struct nexthop *nh) 709{ 710 struct nh_group *nhg; 711 int i; 712 713 ASSERT_RTNL(); 714 nhg = rtnl_dereference(nh->nh_grp); 715 716 info->id = nh->id; 717 info->type = NH_NOTIFIER_INFO_TYPE_GRP_HW_STATS; 718 info->nh_grp_hw_stats = kzalloc(struct_size(info->nh_grp_hw_stats, 719 stats, nhg->num_nh), 720 GFP_KERNEL); 721 if (!info->nh_grp_hw_stats) 722 return -ENOMEM; 723 724 info->nh_grp_hw_stats->num_nh = nhg->num_nh; 725 for (i = 0; i < nhg->num_nh; i++) { 726 struct nh_grp_entry *nhge = &nhg->nh_entries[i]; 727 728 info->nh_grp_hw_stats->stats[i].id = nhge->nh->id; 729 } 730 731 return 0; 732} 733 734static void nh_notifier_grp_hw_stats_fini(struct nh_notifier_info *info) 735{ 736 kfree(info->nh_grp_hw_stats); 737} 738 739void nh_grp_hw_stats_report_delta(struct nh_notifier_grp_hw_stats_info *info, 740 unsigned int nh_idx, 741 u64 delta_packets) 742{ 743 info->hw_stats_used = true; 744 info->stats[nh_idx].packets += delta_packets; 745} 746EXPORT_SYMBOL(nh_grp_hw_stats_report_delta); 747 748static void nh_grp_hw_stats_apply_update(struct nexthop *nh, 749 struct nh_notifier_info *info) 750{ 751 struct nh_group *nhg; 752 int i; 753 754 ASSERT_RTNL(); 755 nhg = rtnl_dereference(nh->nh_grp); 756 757 for (i = 0; i < nhg->num_nh; i++) { 758 struct nh_grp_entry *nhge = &nhg->nh_entries[i]; 759 760 nhge->packets_hw += info->nh_grp_hw_stats->stats[i].packets; 761 } 762} 763 764static int nh_grp_hw_stats_update(struct nexthop *nh, bool *hw_stats_used) 765{ 766 struct nh_notifier_info info = { 767 .net = nh->net, 768 }; 769 struct net *net = nh->net; 770 int err; 771 772 if (nexthop_notifiers_is_empty(net)) { 773 *hw_stats_used = false; 774 return 0; 775 } 776 777 err = nh_notifier_grp_hw_stats_init(&info, nh); 778 if (err) 779 return err; 780 781 err = blocking_notifier_call_chain(&net->nexthop.notifier_chain, 782 NEXTHOP_EVENT_HW_STATS_REPORT_DELTA, 783 &info); 784 785 /* Cache whatever we got, even if there was an error, otherwise the 786 * successful stats retrievals would get lost. 787 */ 788 nh_grp_hw_stats_apply_update(nh, &info); 789 *hw_stats_used = info.nh_grp_hw_stats->hw_stats_used; 790 791 nh_notifier_grp_hw_stats_fini(&info); 792 return notifier_to_errno(err); 793} 794 795static int nla_put_nh_group_stats_entry(struct sk_buff *skb, 796 struct nh_grp_entry *nhge, 797 u32 op_flags) 798{ 799 struct nlattr *nest; 800 u64 packets; 801 802 nh_grp_entry_stats_read(nhge, &packets); 803 804 nest = nla_nest_start(skb, NHA_GROUP_STATS_ENTRY); 805 if (!nest) 806 return -EMSGSIZE; 807 808 if (nla_put_u32(skb, NHA_GROUP_STATS_ENTRY_ID, nhge->nh->id) || 809 nla_put_uint(skb, NHA_GROUP_STATS_ENTRY_PACKETS, 810 packets + nhge->packets_hw)) 811 goto nla_put_failure; 812 813 if (op_flags & NHA_OP_FLAG_DUMP_HW_STATS && 814 nla_put_uint(skb, NHA_GROUP_STATS_ENTRY_PACKETS_HW, 815 nhge->packets_hw)) 816 goto nla_put_failure; 817 818 nla_nest_end(skb, nest); 819 return 0; 820 821nla_put_failure: 822 nla_nest_cancel(skb, nest); 823 return -EMSGSIZE; 824} 825 826static int nla_put_nh_group_stats(struct sk_buff *skb, struct nexthop *nh, 827 u32 op_flags) 828{ 829 struct nh_group *nhg = rtnl_dereference(nh->nh_grp); 830 struct nlattr *nest; 831 bool hw_stats_used; 832 int err; 833 int i; 834 835 if (nla_put_u32(skb, NHA_HW_STATS_ENABLE, nhg->hw_stats)) 836 goto err_out; 837 838 if (op_flags & NHA_OP_FLAG_DUMP_HW_STATS && 839 nhg->hw_stats) { 840 err = nh_grp_hw_stats_update(nh, &hw_stats_used); 841 if (err) 842 goto out; 843 844 if (nla_put_u32(skb, NHA_HW_STATS_USED, hw_stats_used)) 845 goto err_out; 846 } 847 848 nest = nla_nest_start(skb, NHA_GROUP_STATS); 849 if (!nest) 850 goto err_out; 851 852 for (i = 0; i < nhg->num_nh; i++) 853 if (nla_put_nh_group_stats_entry(skb, &nhg->nh_entries[i], 854 op_flags)) 855 goto cancel_out; 856 857 nla_nest_end(skb, nest); 858 return 0; 859 860cancel_out: 861 nla_nest_cancel(skb, nest); 862err_out: 863 err = -EMSGSIZE; 864out: 865 return err; 866} 867 868static int nla_put_nh_group(struct sk_buff *skb, struct nexthop *nh, 869 u32 op_flags, u32 *resp_op_flags) 870{ 871 struct nh_group *nhg = rtnl_dereference(nh->nh_grp); 872 struct nexthop_grp *p; 873 size_t len = nhg->num_nh * sizeof(*p); 874 struct nlattr *nla; 875 u16 group_type = 0; 876 u16 weight; 877 int i; 878 879 *resp_op_flags |= NHA_OP_FLAG_RESP_GRP_RESVD_0; 880 881 if (nhg->hash_threshold) 882 group_type = NEXTHOP_GRP_TYPE_MPATH; 883 else if (nhg->resilient) 884 group_type = NEXTHOP_GRP_TYPE_RES; 885 886 if (nla_put_u16(skb, NHA_GROUP_TYPE, group_type)) 887 goto nla_put_failure; 888 889 nla = nla_reserve(skb, NHA_GROUP, len); 890 if (!nla) 891 goto nla_put_failure; 892 893 p = nla_data(nla); 894 for (i = 0; i < nhg->num_nh; ++i) { 895 weight = nhg->nh_entries[i].weight - 1; 896 897 *p++ = (struct nexthop_grp) { 898 .id = nhg->nh_entries[i].nh->id, 899 .weight = weight, 900 .weight_high = weight >> 8, 901 }; 902 } 903 904 if (nhg->resilient && nla_put_nh_group_res(skb, nhg)) 905 goto nla_put_failure; 906 907 if (op_flags & NHA_OP_FLAG_DUMP_STATS && 908 nla_put_nh_group_stats(skb, nh, op_flags)) 909 goto nla_put_failure; 910 911 return 0; 912 913nla_put_failure: 914 return -EMSGSIZE; 915} 916 917static int nh_fill_node(struct sk_buff *skb, struct nexthop *nh, 918 int event, u32 portid, u32 seq, unsigned int nlflags, 919 u32 op_flags) 920{ 921 struct fib6_nh *fib6_nh; 922 struct fib_nh *fib_nh; 923 struct nlmsghdr *nlh; 924 struct nh_info *nhi; 925 struct nhmsg *nhm; 926 927 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*nhm), nlflags); 928 if (!nlh) 929 return -EMSGSIZE; 930 931 nhm = nlmsg_data(nlh); 932 nhm->nh_family = AF_UNSPEC; 933 nhm->nh_flags = nh->nh_flags; 934 nhm->nh_protocol = nh->protocol; 935 nhm->nh_scope = 0; 936 nhm->resvd = 0; 937 938 if (nla_put_u32(skb, NHA_ID, nh->id)) 939 goto nla_put_failure; 940 941 if (nh->is_group) { 942 struct nh_group *nhg = rtnl_dereference(nh->nh_grp); 943 u32 resp_op_flags = 0; 944 945 if (nhg->fdb_nh && nla_put_flag(skb, NHA_FDB)) 946 goto nla_put_failure; 947 if (nla_put_nh_group(skb, nh, op_flags, &resp_op_flags) || 948 nla_put_u32(skb, NHA_OP_FLAGS, resp_op_flags)) 949 goto nla_put_failure; 950 goto out; 951 } 952 953 nhi = rtnl_dereference(nh->nh_info); 954 nhm->nh_family = nhi->family; 955 if (nhi->reject_nh) { 956 if (nla_put_flag(skb, NHA_BLACKHOLE)) 957 goto nla_put_failure; 958 goto out; 959 } else if (nhi->fdb_nh) { 960 if (nla_put_flag(skb, NHA_FDB)) 961 goto nla_put_failure; 962 } else { 963 const struct net_device *dev; 964 965 dev = nhi->fib_nhc.nhc_dev; 966 if (dev && nla_put_u32(skb, NHA_OIF, dev->ifindex)) 967 goto nla_put_failure; 968 } 969 970 nhm->nh_scope = nhi->fib_nhc.nhc_scope; 971 switch (nhi->family) { 972 case AF_INET: 973 fib_nh = &nhi->fib_nh; 974 if (fib_nh->fib_nh_gw_family && 975 nla_put_be32(skb, NHA_GATEWAY, fib_nh->fib_nh_gw4)) 976 goto nla_put_failure; 977 break; 978 979 case AF_INET6: 980 fib6_nh = &nhi->fib6_nh; 981 if (fib6_nh->fib_nh_gw_family && 982 nla_put_in6_addr(skb, NHA_GATEWAY, &fib6_nh->fib_nh_gw6)) 983 goto nla_put_failure; 984 break; 985 } 986 987 if (lwtunnel_fill_encap(skb, nhi->fib_nhc.nhc_lwtstate, 988 NHA_ENCAP, NHA_ENCAP_TYPE) < 0) 989 goto nla_put_failure; 990 991out: 992 nlmsg_end(skb, nlh); 993 return 0; 994 995nla_put_failure: 996 nlmsg_cancel(skb, nlh); 997 return -EMSGSIZE; 998} 999 1000static size_t nh_nlmsg_size_grp_res(struct nh_group *nhg) 1001{ 1002 return nla_total_size(0) + /* NHA_RES_GROUP */ 1003 nla_total_size(2) + /* NHA_RES_GROUP_BUCKETS */ 1004 nla_total_size(4) + /* NHA_RES_GROUP_IDLE_TIMER */ 1005 nla_total_size(4) + /* NHA_RES_GROUP_UNBALANCED_TIMER */ 1006 nla_total_size_64bit(8);/* NHA_RES_GROUP_UNBALANCED_TIME */ 1007} 1008 1009static size_t nh_nlmsg_size_grp(struct nexthop *nh, u32 op_flags) 1010{ 1011 struct nh_group *nhg = rtnl_dereference(nh->nh_grp); 1012 size_t sz = sizeof(struct nexthop_grp) * nhg->num_nh; 1013 size_t tot = nla_total_size(sz) + 1014 nla_total_size(2) + /* NHA_GROUP_TYPE */ 1015 nla_total_size(0); /* NHA_FDB */ 1016 1017 if (nhg->resilient) 1018 tot += nh_nlmsg_size_grp_res(nhg); 1019 1020 if (op_flags & NHA_OP_FLAG_DUMP_STATS) { 1021 tot += nla_total_size(0) + /* NHA_GROUP_STATS */ 1022 nla_total_size(4); /* NHA_HW_STATS_ENABLE */ 1023 tot += nhg->num_nh * 1024 (nla_total_size(0) + /* NHA_GROUP_STATS_ENTRY */ 1025 nla_total_size(4) + /* NHA_GROUP_STATS_ENTRY_ID */ 1026 nla_total_size_64bit(8)); /* NHA_GROUP_STATS_ENTRY_PACKETS */ 1027 1028 if (op_flags & NHA_OP_FLAG_DUMP_HW_STATS) { 1029 tot += nhg->num_nh * 1030 nla_total_size_64bit(8); /* NHA_GROUP_STATS_ENTRY_PACKETS_HW */ 1031 tot += nla_total_size(4); /* NHA_HW_STATS_USED */ 1032 } 1033 } 1034 1035 return tot; 1036} 1037 1038static size_t nh_nlmsg_size_single(struct nexthop *nh) 1039{ 1040 struct nh_info *nhi = rtnl_dereference(nh->nh_info); 1041 size_t sz; 1042 1043 /* covers NHA_BLACKHOLE since NHA_OIF and BLACKHOLE 1044 * are mutually exclusive 1045 */ 1046 sz = nla_total_size(4); /* NHA_OIF */ 1047 1048 switch (nhi->family) { 1049 case AF_INET: 1050 if (nhi->fib_nh.fib_nh_gw_family) 1051 sz += nla_total_size(4); /* NHA_GATEWAY */ 1052 break; 1053 1054 case AF_INET6: 1055 /* NHA_GATEWAY */ 1056 if (nhi->fib6_nh.fib_nh_gw_family) 1057 sz += nla_total_size(sizeof(const struct in6_addr)); 1058 break; 1059 } 1060 1061 if (nhi->fib_nhc.nhc_lwtstate) { 1062 sz += lwtunnel_get_encap_size(nhi->fib_nhc.nhc_lwtstate); 1063 sz += nla_total_size(2); /* NHA_ENCAP_TYPE */ 1064 } 1065 1066 return sz; 1067} 1068 1069static size_t nh_nlmsg_size(struct nexthop *nh, u32 op_flags) 1070{ 1071 size_t sz = NLMSG_ALIGN(sizeof(struct nhmsg)); 1072 1073 sz += nla_total_size(4); /* NHA_ID */ 1074 1075 if (nh->is_group) 1076 sz += nh_nlmsg_size_grp(nh, op_flags) + 1077 nla_total_size(4) + /* NHA_OP_FLAGS */ 1078 0; 1079 else 1080 sz += nh_nlmsg_size_single(nh); 1081 1082 return sz; 1083} 1084 1085static void nexthop_notify(int event, struct nexthop *nh, struct nl_info *info) 1086{ 1087 unsigned int nlflags = info->nlh ? info->nlh->nlmsg_flags : 0; 1088 u32 seq = info->nlh ? info->nlh->nlmsg_seq : 0; 1089 struct sk_buff *skb; 1090 int err = -ENOBUFS; 1091 1092 skb = nlmsg_new(nh_nlmsg_size(nh, 0), gfp_any()); 1093 if (!skb) 1094 goto errout; 1095 1096 err = nh_fill_node(skb, nh, event, info->portid, seq, nlflags, 0); 1097 if (err < 0) { 1098 /* -EMSGSIZE implies BUG in nh_nlmsg_size() */ 1099 WARN_ON(err == -EMSGSIZE); 1100 kfree_skb(skb); 1101 goto errout; 1102 } 1103 1104 rtnl_notify(skb, info->nl_net, info->portid, RTNLGRP_NEXTHOP, 1105 info->nlh, gfp_any()); 1106 return; 1107errout: 1108 rtnl_set_sk_err(info->nl_net, RTNLGRP_NEXTHOP, err); 1109} 1110 1111static unsigned long nh_res_bucket_used_time(const struct nh_res_bucket *bucket) 1112{ 1113 return (unsigned long)atomic_long_read(&bucket->used_time); 1114} 1115 1116static unsigned long 1117nh_res_bucket_idle_point(const struct nh_res_table *res_table, 1118 const struct nh_res_bucket *bucket, 1119 unsigned long now) 1120{ 1121 unsigned long time = nh_res_bucket_used_time(bucket); 1122 1123 /* Bucket was not used since it was migrated. The idle time is now. */ 1124 if (time == bucket->migrated_time) 1125 return now; 1126 1127 return time + res_table->idle_timer; 1128} 1129 1130static unsigned long 1131nh_res_table_unb_point(const struct nh_res_table *res_table) 1132{ 1133 return res_table->unbalanced_since + res_table->unbalanced_timer; 1134} 1135 1136static void nh_res_bucket_set_idle(const struct nh_res_table *res_table, 1137 struct nh_res_bucket *bucket) 1138{ 1139 unsigned long now = jiffies; 1140 1141 atomic_long_set(&bucket->used_time, (long)now); 1142 bucket->migrated_time = now; 1143} 1144 1145static void nh_res_bucket_set_busy(struct nh_res_bucket *bucket) 1146{ 1147 atomic_long_set(&bucket->used_time, (long)jiffies); 1148} 1149 1150static clock_t nh_res_bucket_idle_time(const struct nh_res_bucket *bucket) 1151{ 1152 unsigned long used_time = nh_res_bucket_used_time(bucket); 1153 1154 return jiffies_delta_to_clock_t(jiffies - used_time); 1155} 1156 1157static int nh_fill_res_bucket(struct sk_buff *skb, struct nexthop *nh, 1158 struct nh_res_bucket *bucket, u16 bucket_index, 1159 int event, u32 portid, u32 seq, 1160 unsigned int nlflags, 1161 struct netlink_ext_ack *extack) 1162{ 1163 struct nh_grp_entry *nhge = nh_res_dereference(bucket->nh_entry); 1164 struct nlmsghdr *nlh; 1165 struct nlattr *nest; 1166 struct nhmsg *nhm; 1167 1168 nlh = nlmsg_put(skb, portid, seq, event, sizeof(*nhm), nlflags); 1169 if (!nlh) 1170 return -EMSGSIZE; 1171 1172 nhm = nlmsg_data(nlh); 1173 nhm->nh_family = AF_UNSPEC; 1174 nhm->nh_flags = bucket->nh_flags; 1175 nhm->nh_protocol = nh->protocol; 1176 nhm->nh_scope = 0; 1177 nhm->resvd = 0; 1178 1179 if (nla_put_u32(skb, NHA_ID, nh->id)) 1180 goto nla_put_failure; 1181 1182 nest = nla_nest_start(skb, NHA_RES_BUCKET); 1183 if (!nest) 1184 goto nla_put_failure; 1185 1186 if (nla_put_u16(skb, NHA_RES_BUCKET_INDEX, bucket_index) || 1187 nla_put_u32(skb, NHA_RES_BUCKET_NH_ID, nhge->nh->id) || 1188 nla_put_u64_64bit(skb, NHA_RES_BUCKET_IDLE_TIME, 1189 nh_res_bucket_idle_time(bucket), 1190 NHA_RES_BUCKET_PAD)) 1191 goto nla_put_failure_nest; 1192 1193 nla_nest_end(skb, nest); 1194 nlmsg_end(skb, nlh); 1195 return 0; 1196 1197nla_put_failure_nest: 1198 nla_nest_cancel(skb, nest); 1199nla_put_failure: 1200 nlmsg_cancel(skb, nlh); 1201 return -EMSGSIZE; 1202} 1203 1204static void nexthop_bucket_notify(struct nh_res_table *res_table, 1205 u16 bucket_index) 1206{ 1207 struct nh_res_bucket *bucket = &res_table->nh_buckets[bucket_index]; 1208 struct nh_grp_entry *nhge = nh_res_dereference(bucket->nh_entry); 1209 struct nexthop *nh = nhge->nh_parent; 1210 struct sk_buff *skb; 1211 int err = -ENOBUFS; 1212 1213 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 1214 if (!skb) 1215 goto errout; 1216 1217 err = nh_fill_res_bucket(skb, nh, bucket, bucket_index, 1218 RTM_NEWNEXTHOPBUCKET, 0, 0, NLM_F_REPLACE, 1219 NULL); 1220 if (err < 0) { 1221 kfree_skb(skb); 1222 goto errout; 1223 } 1224 1225 rtnl_notify(skb, nh->net, 0, RTNLGRP_NEXTHOP, NULL, GFP_KERNEL); 1226 return; 1227errout: 1228 rtnl_set_sk_err(nh->net, RTNLGRP_NEXTHOP, err); 1229} 1230 1231static bool valid_group_nh(struct nexthop *nh, unsigned int npaths, 1232 bool *is_fdb, struct netlink_ext_ack *extack) 1233{ 1234 if (nh->is_group) { 1235 struct nh_group *nhg = rtnl_dereference(nh->nh_grp); 1236 1237 /* Nesting groups within groups is not supported. */ 1238 if (nhg->hash_threshold) { 1239 NL_SET_ERR_MSG(extack, 1240 "Hash-threshold group can not be a nexthop within a group"); 1241 return false; 1242 } 1243 if (nhg->resilient) { 1244 NL_SET_ERR_MSG(extack, 1245 "Resilient group can not be a nexthop within a group"); 1246 return false; 1247 } 1248 *is_fdb = nhg->fdb_nh; 1249 } else { 1250 struct nh_info *nhi = rtnl_dereference(nh->nh_info); 1251 1252 if (nhi->reject_nh && npaths > 1) { 1253 NL_SET_ERR_MSG(extack, 1254 "Blackhole nexthop can not be used in a group with more than 1 path"); 1255 return false; 1256 } 1257 *is_fdb = nhi->fdb_nh; 1258 } 1259 1260 return true; 1261} 1262 1263static int nh_check_attr_fdb_group(struct nexthop *nh, u8 *nh_family, 1264 struct netlink_ext_ack *extack) 1265{ 1266 struct nh_info *nhi; 1267 1268 nhi = rtnl_dereference(nh->nh_info); 1269 1270 if (!nhi->fdb_nh) { 1271 NL_SET_ERR_MSG(extack, "FDB nexthop group can only have fdb nexthops"); 1272 return -EINVAL; 1273 } 1274 1275 if (*nh_family == AF_UNSPEC) { 1276 *nh_family = nhi->family; 1277 } else if (*nh_family != nhi->family) { 1278 NL_SET_ERR_MSG(extack, "FDB nexthop group cannot have mixed family nexthops"); 1279 return -EINVAL; 1280 } 1281 1282 return 0; 1283} 1284 1285static int nh_check_attr_group(struct net *net, 1286 struct nlattr *tb[], size_t tb_size, 1287 u16 nh_grp_type, struct netlink_ext_ack *extack) 1288{ 1289 unsigned int len = nla_len(tb[NHA_GROUP]); 1290 struct nexthop_grp *nhg; 1291 unsigned int i, j; 1292 1293 if (!len || len & (sizeof(struct nexthop_grp) - 1)) { 1294 NL_SET_ERR_MSG(extack, 1295 "Invalid length for nexthop group attribute"); 1296 return -EINVAL; 1297 } 1298 1299 /* convert len to number of nexthop ids */ 1300 len /= sizeof(*nhg); 1301 1302 nhg = nla_data(tb[NHA_GROUP]); 1303 for (i = 0; i < len; ++i) { 1304 if (nhg[i].resvd2) { 1305 NL_SET_ERR_MSG(extack, "Reserved field in nexthop_grp must be 0"); 1306 return -EINVAL; 1307 } 1308 if (nexthop_grp_weight(&nhg[i]) == 0) { 1309 /* 0xffff got passed in, representing weight of 0x10000, 1310 * which is too heavy. 1311 */ 1312 NL_SET_ERR_MSG(extack, "Invalid value for weight"); 1313 return -EINVAL; 1314 } 1315 for (j = i + 1; j < len; ++j) { 1316 if (nhg[i].id == nhg[j].id) { 1317 NL_SET_ERR_MSG(extack, "Nexthop id can not be used twice in a group"); 1318 return -EINVAL; 1319 } 1320 } 1321 } 1322 1323 nhg = nla_data(tb[NHA_GROUP]); 1324 for (i = NHA_GROUP_TYPE + 1; i < tb_size; ++i) { 1325 if (!tb[i]) 1326 continue; 1327 switch (i) { 1328 case NHA_HW_STATS_ENABLE: 1329 case NHA_FDB: 1330 continue; 1331 case NHA_RES_GROUP: 1332 if (nh_grp_type == NEXTHOP_GRP_TYPE_RES) 1333 continue; 1334 break; 1335 } 1336 NL_SET_ERR_MSG(extack, 1337 "No other attributes can be set in nexthop groups"); 1338 return -EINVAL; 1339 } 1340 1341 return 0; 1342} 1343 1344static int nh_check_attr_group_rtnl(struct net *net, struct nlattr *tb[], 1345 struct netlink_ext_ack *extack) 1346{ 1347 u8 nh_family = AF_UNSPEC; 1348 struct nexthop_grp *nhg; 1349 unsigned int len; 1350 unsigned int i; 1351 u8 nhg_fdb; 1352 1353 len = nla_len(tb[NHA_GROUP]) / sizeof(*nhg); 1354 nhg = nla_data(tb[NHA_GROUP]); 1355 nhg_fdb = !!tb[NHA_FDB]; 1356 1357 for (i = 0; i < len; i++) { 1358 struct nexthop *nh; 1359 bool is_fdb_nh; 1360 1361 nh = nexthop_find_by_id(net, nhg[i].id); 1362 if (!nh) { 1363 NL_SET_ERR_MSG(extack, "Invalid nexthop id"); 1364 return -EINVAL; 1365 } 1366 if (!valid_group_nh(nh, len, &is_fdb_nh, extack)) 1367 return -EINVAL; 1368 1369 if (nhg_fdb && nh_check_attr_fdb_group(nh, &nh_family, extack)) 1370 return -EINVAL; 1371 1372 if (!nhg_fdb && is_fdb_nh) { 1373 NL_SET_ERR_MSG(extack, "Non FDB nexthop group cannot have fdb nexthops"); 1374 return -EINVAL; 1375 } 1376 } 1377 1378 return 0; 1379} 1380 1381static bool ipv6_good_nh(const struct fib6_nh *nh) 1382{ 1383 int state = NUD_REACHABLE; 1384 struct neighbour *n; 1385 1386 rcu_read_lock(); 1387 1388 n = __ipv6_neigh_lookup_noref_stub(nh->fib_nh_dev, &nh->fib_nh_gw6); 1389 if (n) 1390 state = READ_ONCE(n->nud_state); 1391 1392 rcu_read_unlock(); 1393 1394 return !!(state & NUD_VALID); 1395} 1396 1397static bool ipv4_good_nh(const struct fib_nh *nh) 1398{ 1399 int state = NUD_REACHABLE; 1400 struct neighbour *n; 1401 1402 rcu_read_lock(); 1403 1404 n = __ipv4_neigh_lookup_noref(nh->fib_nh_dev, 1405 (__force u32)nh->fib_nh_gw4); 1406 if (n) 1407 state = READ_ONCE(n->nud_state); 1408 1409 rcu_read_unlock(); 1410 1411 return !!(state & NUD_VALID); 1412} 1413 1414static bool nexthop_is_good_nh(const struct nexthop *nh) 1415{ 1416 struct nh_info *nhi = rcu_dereference(nh->nh_info); 1417 1418 switch (nhi->family) { 1419 case AF_INET: 1420 return ipv4_good_nh(&nhi->fib_nh); 1421 case AF_INET6: 1422 return ipv6_good_nh(&nhi->fib6_nh); 1423 } 1424 1425 return false; 1426} 1427 1428static struct nexthop *nexthop_select_path_fdb(struct nh_group *nhg, int hash) 1429{ 1430 int i; 1431 1432 for (i = 0; i < nhg->num_nh; i++) { 1433 struct nh_grp_entry *nhge = &nhg->nh_entries[i]; 1434 1435 if (hash > atomic_read(&nhge->hthr.upper_bound)) 1436 continue; 1437 1438 nh_grp_entry_stats_inc(nhge); 1439 return nhge->nh; 1440 } 1441 1442 WARN_ON_ONCE(1); 1443 return NULL; 1444} 1445 1446static struct nexthop *nexthop_select_path_hthr(struct nh_group *nhg, int hash) 1447{ 1448 struct nh_grp_entry *nhge0 = NULL; 1449 int i; 1450 1451 if (nhg->fdb_nh) 1452 return nexthop_select_path_fdb(nhg, hash); 1453 1454 for (i = 0; i < nhg->num_nh; ++i) { 1455 struct nh_grp_entry *nhge = &nhg->nh_entries[i]; 1456 1457 /* nexthops always check if it is good and does 1458 * not rely on a sysctl for this behavior 1459 */ 1460 if (!nexthop_is_good_nh(nhge->nh)) 1461 continue; 1462 1463 if (!nhge0) 1464 nhge0 = nhge; 1465 1466 if (hash > atomic_read(&nhge->hthr.upper_bound)) 1467 continue; 1468 1469 nh_grp_entry_stats_inc(nhge); 1470 return nhge->nh; 1471 } 1472 1473 if (!nhge0) 1474 nhge0 = &nhg->nh_entries[0]; 1475 nh_grp_entry_stats_inc(nhge0); 1476 return nhge0->nh; 1477} 1478 1479static struct nexthop *nexthop_select_path_res(struct nh_group *nhg, int hash) 1480{ 1481 struct nh_res_table *res_table = rcu_dereference(nhg->res_table); 1482 u16 bucket_index = hash % res_table->num_nh_buckets; 1483 struct nh_res_bucket *bucket; 1484 struct nh_grp_entry *nhge; 1485 1486 /* nexthop_select_path() is expected to return a non-NULL value, so 1487 * skip protocol validation and just hand out whatever there is. 1488 */ 1489 bucket = &res_table->nh_buckets[bucket_index]; 1490 nh_res_bucket_set_busy(bucket); 1491 nhge = rcu_dereference(bucket->nh_entry); 1492 nh_grp_entry_stats_inc(nhge); 1493 return nhge->nh; 1494} 1495 1496struct nexthop *nexthop_select_path(struct nexthop *nh, int hash) 1497{ 1498 struct nh_group *nhg; 1499 1500 if (!nh->is_group) 1501 return nh; 1502 1503 nhg = rcu_dereference(nh->nh_grp); 1504 if (nhg->hash_threshold) 1505 return nexthop_select_path_hthr(nhg, hash); 1506 else if (nhg->resilient) 1507 return nexthop_select_path_res(nhg, hash); 1508 1509 /* Unreachable. */ 1510 return NULL; 1511} 1512EXPORT_SYMBOL_GPL(nexthop_select_path); 1513 1514int nexthop_for_each_fib6_nh(struct nexthop *nh, 1515 int (*cb)(struct fib6_nh *nh, void *arg), 1516 void *arg) 1517{ 1518 struct nh_info *nhi; 1519 int err; 1520 1521 if (nh->is_group) { 1522 struct nh_group *nhg; 1523 int i; 1524 1525 nhg = rcu_dereference_rtnl(nh->nh_grp); 1526 for (i = 0; i < nhg->num_nh; i++) { 1527 struct nh_grp_entry *nhge = &nhg->nh_entries[i]; 1528 1529 nhi = rcu_dereference_rtnl(nhge->nh->nh_info); 1530 err = cb(&nhi->fib6_nh, arg); 1531 if (err) 1532 return err; 1533 } 1534 } else { 1535 nhi = rcu_dereference_rtnl(nh->nh_info); 1536 err = cb(&nhi->fib6_nh, arg); 1537 if (err) 1538 return err; 1539 } 1540 1541 return 0; 1542} 1543EXPORT_SYMBOL_GPL(nexthop_for_each_fib6_nh); 1544 1545static int check_src_addr(const struct in6_addr *saddr, 1546 struct netlink_ext_ack *extack) 1547{ 1548 if (!ipv6_addr_any(saddr)) { 1549 NL_SET_ERR_MSG(extack, "IPv6 routes using source address can not use nexthop objects"); 1550 return -EINVAL; 1551 } 1552 return 0; 1553} 1554 1555int fib6_check_nexthop(struct nexthop *nh, struct fib6_config *cfg, 1556 struct netlink_ext_ack *extack) 1557{ 1558 struct nh_info *nhi; 1559 bool is_fdb_nh; 1560 1561 /* fib6_src is unique to a fib6_info and limits the ability to cache 1562 * routes in fib6_nh within a nexthop that is potentially shared 1563 * across multiple fib entries. If the config wants to use source 1564 * routing it can not use nexthop objects. mlxsw also does not allow 1565 * fib6_src on routes. 1566 */ 1567 if (cfg && check_src_addr(&cfg->fc_src, extack) < 0) 1568 return -EINVAL; 1569 1570 if (nh->is_group) { 1571 struct nh_group *nhg; 1572 1573 nhg = rcu_dereference_rtnl(nh->nh_grp); 1574 if (nhg->has_v4) 1575 goto no_v4_nh; 1576 is_fdb_nh = nhg->fdb_nh; 1577 } else { 1578 nhi = rcu_dereference_rtnl(nh->nh_info); 1579 if (nhi->family == AF_INET) 1580 goto no_v4_nh; 1581 is_fdb_nh = nhi->fdb_nh; 1582 } 1583 1584 if (is_fdb_nh) { 1585 NL_SET_ERR_MSG(extack, "Route cannot point to a fdb nexthop"); 1586 return -EINVAL; 1587 } 1588 1589 return 0; 1590no_v4_nh: 1591 NL_SET_ERR_MSG(extack, "IPv6 routes can not use an IPv4 nexthop"); 1592 return -EINVAL; 1593} 1594EXPORT_SYMBOL_GPL(fib6_check_nexthop); 1595 1596/* if existing nexthop has ipv6 routes linked to it, need 1597 * to verify this new spec works with ipv6 1598 */ 1599static int fib6_check_nh_list(struct nexthop *old, struct nexthop *new, 1600 struct netlink_ext_ack *extack) 1601{ 1602 struct fib6_info *f6i; 1603 1604 if (list_empty(&old->f6i_list)) 1605 return 0; 1606 1607 list_for_each_entry(f6i, &old->f6i_list, nh_list) { 1608 if (check_src_addr(&f6i->fib6_src.addr, extack) < 0) 1609 return -EINVAL; 1610 } 1611 1612 return fib6_check_nexthop(new, NULL, extack); 1613} 1614 1615static int nexthop_check_scope(struct nh_info *nhi, u8 scope, 1616 struct netlink_ext_ack *extack) 1617{ 1618 if (scope == RT_SCOPE_HOST && nhi->fib_nhc.nhc_gw_family) { 1619 NL_SET_ERR_MSG(extack, 1620 "Route with host scope can not have a gateway"); 1621 return -EINVAL; 1622 } 1623 1624 if (nhi->fib_nhc.nhc_flags & RTNH_F_ONLINK && scope >= RT_SCOPE_LINK) { 1625 NL_SET_ERR_MSG(extack, "Scope mismatch with nexthop"); 1626 return -EINVAL; 1627 } 1628 1629 return 0; 1630} 1631 1632/* Invoked by fib add code to verify nexthop by id is ok with 1633 * config for prefix; parts of fib_check_nh not done when nexthop 1634 * object is used. 1635 */ 1636int fib_check_nexthop(struct nexthop *nh, u8 scope, 1637 struct netlink_ext_ack *extack) 1638{ 1639 struct nh_info *nhi; 1640 int err = 0; 1641 1642 if (nh->is_group) { 1643 struct nh_group *nhg; 1644 1645 nhg = rtnl_dereference(nh->nh_grp); 1646 if (nhg->fdb_nh) { 1647 NL_SET_ERR_MSG(extack, "Route cannot point to a fdb nexthop"); 1648 err = -EINVAL; 1649 goto out; 1650 } 1651 1652 if (scope == RT_SCOPE_HOST) { 1653 NL_SET_ERR_MSG(extack, "Route with host scope can not have multiple nexthops"); 1654 err = -EINVAL; 1655 goto out; 1656 } 1657 1658 /* all nexthops in a group have the same scope */ 1659 nhi = rtnl_dereference(nhg->nh_entries[0].nh->nh_info); 1660 err = nexthop_check_scope(nhi, scope, extack); 1661 } else { 1662 nhi = rtnl_dereference(nh->nh_info); 1663 if (nhi->fdb_nh) { 1664 NL_SET_ERR_MSG(extack, "Route cannot point to a fdb nexthop"); 1665 err = -EINVAL; 1666 goto out; 1667 } 1668 err = nexthop_check_scope(nhi, scope, extack); 1669 } 1670 1671out: 1672 return err; 1673} 1674 1675static int fib_check_nh_list(struct nexthop *old, struct nexthop *new, 1676 struct netlink_ext_ack *extack) 1677{ 1678 struct fib_info *fi; 1679 1680 list_for_each_entry(fi, &old->fi_list, nh_list) { 1681 int err; 1682 1683 err = fib_check_nexthop(new, fi->fib_scope, extack); 1684 if (err) 1685 return err; 1686 } 1687 return 0; 1688} 1689 1690static bool nh_res_nhge_is_balanced(const struct nh_grp_entry *nhge) 1691{ 1692 return nhge->res.count_buckets == nhge->res.wants_buckets; 1693} 1694 1695static bool nh_res_nhge_is_ow(const struct nh_grp_entry *nhge) 1696{ 1697 return nhge->res.count_buckets > nhge->res.wants_buckets; 1698} 1699 1700static bool nh_res_nhge_is_uw(const struct nh_grp_entry *nhge) 1701{ 1702 return nhge->res.count_buckets < nhge->res.wants_buckets; 1703} 1704 1705static bool nh_res_table_is_balanced(const struct nh_res_table *res_table) 1706{ 1707 return list_empty(&res_table->uw_nh_entries); 1708} 1709 1710static void nh_res_bucket_unset_nh(struct nh_res_bucket *bucket) 1711{ 1712 struct nh_grp_entry *nhge; 1713 1714 if (bucket->occupied) { 1715 nhge = nh_res_dereference(bucket->nh_entry); 1716 nhge->res.count_buckets--; 1717 bucket->occupied = false; 1718 } 1719} 1720 1721static void nh_res_bucket_set_nh(struct nh_res_bucket *bucket, 1722 struct nh_grp_entry *nhge) 1723{ 1724 nh_res_bucket_unset_nh(bucket); 1725 1726 bucket->occupied = true; 1727 rcu_assign_pointer(bucket->nh_entry, nhge); 1728 nhge->res.count_buckets++; 1729} 1730 1731static bool nh_res_bucket_should_migrate(struct nh_res_table *res_table, 1732 struct nh_res_bucket *bucket, 1733 unsigned long *deadline, bool *force) 1734{ 1735 unsigned long now = jiffies; 1736 struct nh_grp_entry *nhge; 1737 unsigned long idle_point; 1738 1739 if (!bucket->occupied) { 1740 /* The bucket is not occupied, its NHGE pointer is either 1741 * NULL or obsolete. We _have to_ migrate: set force. 1742 */ 1743 *force = true; 1744 return true; 1745 } 1746 1747 nhge = nh_res_dereference(bucket->nh_entry); 1748 1749 /* If the bucket is populated by an underweight or balanced 1750 * nexthop, do not migrate. 1751 */ 1752 if (!nh_res_nhge_is_ow(nhge)) 1753 return false; 1754 1755 /* At this point we know that the bucket is populated with an 1756 * overweight nexthop. It needs to be migrated to a new nexthop if 1757 * the idle timer of unbalanced timer expired. 1758 */ 1759 1760 idle_point = nh_res_bucket_idle_point(res_table, bucket, now); 1761 if (time_after_eq(now, idle_point)) { 1762 /* The bucket is idle. We _can_ migrate: unset force. */ 1763 *force = false; 1764 return true; 1765 } 1766 1767 /* Unbalanced timer of 0 means "never force". */ 1768 if (res_table->unbalanced_timer) { 1769 unsigned long unb_point; 1770 1771 unb_point = nh_res_table_unb_point(res_table); 1772 if (time_after(now, unb_point)) { 1773 /* The bucket is not idle, but the unbalanced timer 1774 * expired. We _can_ migrate, but set force anyway, 1775 * so that drivers know to ignore activity reports 1776 * from the HW. 1777 */ 1778 *force = true; 1779 return true; 1780 } 1781 1782 nh_res_time_set_deadline(unb_point, deadline); 1783 } 1784 1785 nh_res_time_set_deadline(idle_point, deadline); 1786 return false; 1787} 1788 1789static bool nh_res_bucket_migrate(struct nh_res_table *res_table, 1790 u16 bucket_index, bool notify, 1791 bool notify_nl, bool force) 1792{ 1793 struct nh_res_bucket *bucket = &res_table->nh_buckets[bucket_index]; 1794 struct nh_grp_entry *new_nhge; 1795 struct netlink_ext_ack extack; 1796 int err; 1797 1798 new_nhge = list_first_entry_or_null(&res_table->uw_nh_entries, 1799 struct nh_grp_entry, 1800 res.uw_nh_entry); 1801 if (WARN_ON_ONCE(!new_nhge)) 1802 /* If this function is called, "bucket" is either not 1803 * occupied, or it belongs to a next hop that is 1804 * overweight. In either case, there ought to be a 1805 * corresponding underweight next hop. 1806 */ 1807 return false; 1808 1809 if (notify) { 1810 struct nh_grp_entry *old_nhge; 1811 1812 old_nhge = nh_res_dereference(bucket->nh_entry); 1813 err = call_nexthop_res_bucket_notifiers(res_table->net, 1814 res_table->nhg_id, 1815 bucket_index, force, 1816 old_nhge->nh, 1817 new_nhge->nh, &extack); 1818 if (err) { 1819 pr_err_ratelimited("%s\n", extack._msg); 1820 if (!force) 1821 return false; 1822 /* It is not possible to veto a forced replacement, so 1823 * just clear the hardware flags from the nexthop 1824 * bucket to indicate to user space that this bucket is 1825 * not correctly populated in hardware. 1826 */ 1827 bucket->nh_flags &= ~(RTNH_F_OFFLOAD | RTNH_F_TRAP); 1828 } 1829 } 1830 1831 nh_res_bucket_set_nh(bucket, new_nhge); 1832 nh_res_bucket_set_idle(res_table, bucket); 1833 1834 if (notify_nl) 1835 nexthop_bucket_notify(res_table, bucket_index); 1836 1837 if (nh_res_nhge_is_balanced(new_nhge)) 1838 list_del(&new_nhge->res.uw_nh_entry); 1839 return true; 1840} 1841 1842#define NH_RES_UPKEEP_DW_MINIMUM_INTERVAL (HZ / 2) 1843 1844static void nh_res_table_upkeep(struct nh_res_table *res_table, 1845 bool notify, bool notify_nl) 1846{ 1847 unsigned long now = jiffies; 1848 unsigned long deadline; 1849 u16 i; 1850 1851 /* Deadline is the next time that upkeep should be run. It is the 1852 * earliest time at which one of the buckets might be migrated. 1853 * Start at the most pessimistic estimate: either unbalanced_timer 1854 * from now, or if there is none, idle_timer from now. For each 1855 * encountered time point, call nh_res_time_set_deadline() to 1856 * refine the estimate. 1857 */ 1858 if (res_table->unbalanced_timer) 1859 deadline = now + res_table->unbalanced_timer; 1860 else 1861 deadline = now + res_table->idle_timer; 1862 1863 for (i = 0; i < res_table->num_nh_buckets; i++) { 1864 struct nh_res_bucket *bucket = &res_table->nh_buckets[i]; 1865 bool force; 1866 1867 if (nh_res_bucket_should_migrate(res_table, bucket, 1868 &deadline, &force)) { 1869 if (!nh_res_bucket_migrate(res_table, i, notify, 1870 notify_nl, force)) { 1871 unsigned long idle_point; 1872 1873 /* A driver can override the migration 1874 * decision if the HW reports that the 1875 * bucket is actually not idle. Therefore 1876 * remark the bucket as busy again and 1877 * update the deadline. 1878 */ 1879 nh_res_bucket_set_busy(bucket); 1880 idle_point = nh_res_bucket_idle_point(res_table, 1881 bucket, 1882 now); 1883 nh_res_time_set_deadline(idle_point, &deadline); 1884 } 1885 } 1886 } 1887 1888 /* If the group is still unbalanced, schedule the next upkeep to 1889 * either the deadline computed above, or the minimum deadline, 1890 * whichever comes later. 1891 */ 1892 if (!nh_res_table_is_balanced(res_table)) { 1893 unsigned long now = jiffies; 1894 unsigned long min_deadline; 1895 1896 min_deadline = now + NH_RES_UPKEEP_DW_MINIMUM_INTERVAL; 1897 if (time_before(deadline, min_deadline)) 1898 deadline = min_deadline; 1899 1900 queue_delayed_work(system_power_efficient_wq, 1901 &res_table->upkeep_dw, deadline - now); 1902 } 1903} 1904 1905static void nh_res_table_upkeep_dw(struct work_struct *work) 1906{ 1907 struct delayed_work *dw = to_delayed_work(work); 1908 struct nh_res_table *res_table; 1909 1910 res_table = container_of(dw, struct nh_res_table, upkeep_dw); 1911 nh_res_table_upkeep(res_table, true, true); 1912} 1913 1914static void nh_res_table_cancel_upkeep(struct nh_res_table *res_table) 1915{ 1916 cancel_delayed_work_sync(&res_table->upkeep_dw); 1917} 1918 1919static void nh_res_group_rebalance(struct nh_group *nhg, 1920 struct nh_res_table *res_table) 1921{ 1922 u16 prev_upper_bound = 0; 1923 u32 total = 0; 1924 u32 w = 0; 1925 int i; 1926 1927 INIT_LIST_HEAD(&res_table->uw_nh_entries); 1928 1929 for (i = 0; i < nhg->num_nh; ++i) 1930 total += nhg->nh_entries[i].weight; 1931 1932 for (i = 0; i < nhg->num_nh; ++i) { 1933 struct nh_grp_entry *nhge = &nhg->nh_entries[i]; 1934 u16 upper_bound; 1935 u64 btw; 1936 1937 w += nhge->weight; 1938 btw = ((u64)res_table->num_nh_buckets) * w; 1939 upper_bound = DIV_ROUND_CLOSEST_ULL(btw, total); 1940 nhge->res.wants_buckets = upper_bound - prev_upper_bound; 1941 prev_upper_bound = upper_bound; 1942 1943 if (nh_res_nhge_is_uw(nhge)) { 1944 if (list_empty(&res_table->uw_nh_entries)) 1945 res_table->unbalanced_since = jiffies; 1946 list_add(&nhge->res.uw_nh_entry, 1947 &res_table->uw_nh_entries); 1948 } 1949 } 1950} 1951 1952/* Migrate buckets in res_table so that they reference NHGE's from NHG with 1953 * the right NH ID. Set those buckets that do not have a corresponding NHGE 1954 * entry in NHG as not occupied. 1955 */ 1956static void nh_res_table_migrate_buckets(struct nh_res_table *res_table, 1957 struct nh_group *nhg) 1958{ 1959 u16 i; 1960 1961 for (i = 0; i < res_table->num_nh_buckets; i++) { 1962 struct nh_res_bucket *bucket = &res_table->nh_buckets[i]; 1963 u32 id = rtnl_dereference(bucket->nh_entry)->nh->id; 1964 bool found = false; 1965 int j; 1966 1967 for (j = 0; j < nhg->num_nh; j++) { 1968 struct nh_grp_entry *nhge = &nhg->nh_entries[j]; 1969 1970 if (nhge->nh->id == id) { 1971 nh_res_bucket_set_nh(bucket, nhge); 1972 found = true; 1973 break; 1974 } 1975 } 1976 1977 if (!found) 1978 nh_res_bucket_unset_nh(bucket); 1979 } 1980} 1981 1982static void replace_nexthop_grp_res(struct nh_group *oldg, 1983 struct nh_group *newg) 1984{ 1985 /* For NH group replacement, the new NHG might only have a stub 1986 * hash table with 0 buckets, because the number of buckets was not 1987 * specified. For NH removal, oldg and newg both reference the same 1988 * res_table. So in any case, in the following, we want to work 1989 * with oldg->res_table. 1990 */ 1991 struct nh_res_table *old_res_table = rtnl_dereference(oldg->res_table); 1992 unsigned long prev_unbalanced_since = old_res_table->unbalanced_since; 1993 bool prev_has_uw = !list_empty(&old_res_table->uw_nh_entries); 1994 1995 nh_res_table_cancel_upkeep(old_res_table); 1996 nh_res_table_migrate_buckets(old_res_table, newg); 1997 nh_res_group_rebalance(newg, old_res_table); 1998 if (prev_has_uw && !list_empty(&old_res_table->uw_nh_entries)) 1999 old_res_table->unbalanced_since = prev_unbalanced_since; 2000 nh_res_table_upkeep(old_res_table, true, false); 2001} 2002 2003static void nh_hthr_group_rebalance(struct nh_group *nhg) 2004{ 2005 u32 total = 0; 2006 u32 w = 0; 2007 int i; 2008 2009 for (i = 0; i < nhg->num_nh; ++i) 2010 total += nhg->nh_entries[i].weight; 2011 2012 for (i = 0; i < nhg->num_nh; ++i) { 2013 struct nh_grp_entry *nhge = &nhg->nh_entries[i]; 2014 u32 upper_bound; 2015 2016 w += nhge->weight; 2017 upper_bound = DIV_ROUND_CLOSEST_ULL((u64)w << 31, total) - 1; 2018 atomic_set(&nhge->hthr.upper_bound, upper_bound); 2019 } 2020} 2021 2022static void remove_nh_grp_entry(struct net *net, struct nh_grp_entry *nhge, 2023 struct nl_info *nlinfo, 2024 struct list_head *deferred_free) 2025{ 2026 struct nh_grp_entry *nhges, *new_nhges; 2027 struct nexthop *nhp = nhge->nh_parent; 2028 struct netlink_ext_ack extack; 2029 struct nexthop *nh = nhge->nh; 2030 struct nh_group *nhg, *newg; 2031 int i, j, err; 2032 2033 WARN_ON(!nh); 2034 2035 nhg = rtnl_dereference(nhp->nh_grp); 2036 newg = nhg->spare; 2037 2038 /* last entry, keep it visible and remove the parent */ 2039 if (nhg->num_nh == 1) { 2040 remove_nexthop(net, nhp, nlinfo); 2041 return; 2042 } 2043 2044 newg->has_v4 = false; 2045 newg->is_multipath = nhg->is_multipath; 2046 newg->hash_threshold = nhg->hash_threshold; 2047 newg->resilient = nhg->resilient; 2048 newg->fdb_nh = nhg->fdb_nh; 2049 newg->num_nh = nhg->num_nh; 2050 2051 /* copy old entries to new except the one getting removed */ 2052 nhges = nhg->nh_entries; 2053 new_nhges = newg->nh_entries; 2054 for (i = 0, j = 0; i < nhg->num_nh; ++i) { 2055 struct nh_info *nhi; 2056 2057 /* current nexthop getting removed */ 2058 if (nhg->nh_entries[i].nh == nh) { 2059 newg->num_nh--; 2060 continue; 2061 } 2062 2063 nhi = rtnl_dereference(nhges[i].nh->nh_info); 2064 if (nhi->family == AF_INET) 2065 newg->has_v4 = true; 2066 2067 list_del(&nhges[i].nh_list); 2068 new_nhges[j].stats = nhges[i].stats; 2069 new_nhges[j].nh_parent = nhges[i].nh_parent; 2070 new_nhges[j].nh = nhges[i].nh; 2071 new_nhges[j].weight = nhges[i].weight; 2072 list_add(&new_nhges[j].nh_list, &new_nhges[j].nh->grp_list); 2073 j++; 2074 } 2075 2076 if (newg->hash_threshold) 2077 nh_hthr_group_rebalance(newg); 2078 else if (newg->resilient) 2079 replace_nexthop_grp_res(nhg, newg); 2080 2081 rcu_assign_pointer(nhp->nh_grp, newg); 2082 2083 list_del(&nhge->nh_list); 2084 nexthop_put(nhge->nh); 2085 list_add(&nhge->nh_list, deferred_free); 2086 2087 /* Removal of a NH from a resilient group is notified through 2088 * bucket notifications. 2089 */ 2090 if (newg->hash_threshold) { 2091 err = call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, nhp, 2092 &extack); 2093 if (err) 2094 pr_err("%s\n", extack._msg); 2095 } 2096 2097 if (nlinfo) 2098 nexthop_notify(RTM_NEWNEXTHOP, nhp, nlinfo); 2099} 2100 2101static void remove_nexthop_from_groups(struct net *net, struct nexthop *nh, 2102 struct nl_info *nlinfo) 2103{ 2104 struct nh_grp_entry *nhge, *tmp; 2105 LIST_HEAD(deferred_free); 2106 2107 /* If there is nothing to do, let's avoid the costly call to 2108 * synchronize_net() 2109 */ 2110 if (list_empty(&nh->grp_list)) 2111 return; 2112 2113 list_for_each_entry_safe(nhge, tmp, &nh->grp_list, nh_list) 2114 remove_nh_grp_entry(net, nhge, nlinfo, &deferred_free); 2115 2116 /* make sure all see the newly published array before releasing rtnl */ 2117 synchronize_net(); 2118 2119 /* Now safe to free percpu stats — all RCU readers have finished */ 2120 list_for_each_entry_safe(nhge, tmp, &deferred_free, nh_list) { 2121 list_del(&nhge->nh_list); 2122 free_percpu(nhge->stats); 2123 } 2124} 2125 2126static void remove_nexthop_group(struct nexthop *nh, struct nl_info *nlinfo) 2127{ 2128 struct nh_group *nhg = rcu_dereference_rtnl(nh->nh_grp); 2129 struct nh_res_table *res_table; 2130 int i, num_nh = nhg->num_nh; 2131 2132 for (i = 0; i < num_nh; ++i) { 2133 struct nh_grp_entry *nhge = &nhg->nh_entries[i]; 2134 2135 if (WARN_ON(!nhge->nh)) 2136 continue; 2137 2138 list_del_init(&nhge->nh_list); 2139 } 2140 2141 if (nhg->resilient) { 2142 res_table = rtnl_dereference(nhg->res_table); 2143 nh_res_table_cancel_upkeep(res_table); 2144 } 2145} 2146 2147/* not called for nexthop replace */ 2148static void __remove_nexthop_fib(struct net *net, struct nexthop *nh) 2149{ 2150 struct fib6_info *f6i; 2151 bool do_flush = false; 2152 struct fib_info *fi; 2153 2154 list_for_each_entry(fi, &nh->fi_list, nh_list) { 2155 fi->fib_flags |= RTNH_F_DEAD; 2156 do_flush = true; 2157 } 2158 if (do_flush) 2159 fib_flush(net); 2160 2161 spin_lock_bh(&nh->lock); 2162 2163 nh->dead = true; 2164 2165 while (!list_empty(&nh->f6i_list)) { 2166 f6i = list_first_entry(&nh->f6i_list, typeof(*f6i), nh_list); 2167 2168 /* __ip6_del_rt does a release, so do a hold here */ 2169 fib6_info_hold(f6i); 2170 2171 spin_unlock_bh(&nh->lock); 2172 ipv6_stub->ip6_del_rt(net, f6i, 2173 !READ_ONCE(net->ipv4.sysctl_nexthop_compat_mode)); 2174 2175 spin_lock_bh(&nh->lock); 2176 } 2177 2178 spin_unlock_bh(&nh->lock); 2179} 2180 2181static void __remove_nexthop(struct net *net, struct nexthop *nh, 2182 struct nl_info *nlinfo) 2183{ 2184 __remove_nexthop_fib(net, nh); 2185 2186 if (nh->is_group) { 2187 remove_nexthop_group(nh, nlinfo); 2188 } else { 2189 struct nh_info *nhi; 2190 2191 nhi = rtnl_dereference(nh->nh_info); 2192 if (nhi->fib_nhc.nhc_dev) 2193 hlist_del(&nhi->dev_hash); 2194 2195 remove_nexthop_from_groups(net, nh, nlinfo); 2196 } 2197} 2198 2199static void remove_nexthop(struct net *net, struct nexthop *nh, 2200 struct nl_info *nlinfo) 2201{ 2202 call_nexthop_notifiers(net, NEXTHOP_EVENT_DEL, nh, NULL); 2203 2204 /* remove from the tree */ 2205 rb_erase(&nh->rb_node, &net->nexthop.rb_root); 2206 2207 if (nlinfo) 2208 nexthop_notify(RTM_DELNEXTHOP, nh, nlinfo); 2209 2210 __remove_nexthop(net, nh, nlinfo); 2211 nh_base_seq_inc(net); 2212 2213 nexthop_put(nh); 2214} 2215 2216/* if any FIB entries reference this nexthop, any dst entries 2217 * need to be regenerated 2218 */ 2219static void nh_rt_cache_flush(struct net *net, struct nexthop *nh, 2220 struct nexthop *replaced_nh) 2221{ 2222 struct fib6_info *f6i; 2223 struct nh_group *nhg; 2224 int i; 2225 2226 if (!list_empty(&nh->fi_list)) 2227 rt_cache_flush(net); 2228 2229 list_for_each_entry(f6i, &nh->f6i_list, nh_list) 2230 ipv6_stub->fib6_update_sernum(net, f6i); 2231 2232 /* if an IPv6 group was replaced, we have to release all old 2233 * dsts to make sure all refcounts are released 2234 */ 2235 if (!replaced_nh->is_group) 2236 return; 2237 2238 nhg = rtnl_dereference(replaced_nh->nh_grp); 2239 for (i = 0; i < nhg->num_nh; i++) { 2240 struct nh_grp_entry *nhge = &nhg->nh_entries[i]; 2241 struct nh_info *nhi = rtnl_dereference(nhge->nh->nh_info); 2242 2243 if (nhi->family == AF_INET6) 2244 ipv6_stub->fib6_nh_release_dsts(&nhi->fib6_nh); 2245 } 2246} 2247 2248static int replace_nexthop_grp(struct net *net, struct nexthop *old, 2249 struct nexthop *new, const struct nh_config *cfg, 2250 struct netlink_ext_ack *extack) 2251{ 2252 struct nh_res_table *tmp_table = NULL; 2253 struct nh_res_table *new_res_table; 2254 struct nh_res_table *old_res_table; 2255 struct nh_group *oldg, *newg; 2256 int i, err; 2257 2258 if (!new->is_group) { 2259 NL_SET_ERR_MSG(extack, "Can not replace a nexthop group with a nexthop."); 2260 return -EINVAL; 2261 } 2262 2263 oldg = rtnl_dereference(old->nh_grp); 2264 newg = rtnl_dereference(new->nh_grp); 2265 2266 if (newg->hash_threshold != oldg->hash_threshold) { 2267 NL_SET_ERR_MSG(extack, "Can not replace a nexthop group with one of a different type."); 2268 return -EINVAL; 2269 } 2270 2271 if (newg->hash_threshold) { 2272 err = call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, new, 2273 extack); 2274 if (err) 2275 return err; 2276 } else if (newg->resilient) { 2277 new_res_table = rtnl_dereference(newg->res_table); 2278 old_res_table = rtnl_dereference(oldg->res_table); 2279 2280 /* Accept if num_nh_buckets was not given, but if it was 2281 * given, demand that the value be correct. 2282 */ 2283 if (cfg->nh_grp_res_has_num_buckets && 2284 cfg->nh_grp_res_num_buckets != 2285 old_res_table->num_nh_buckets) { 2286 NL_SET_ERR_MSG(extack, "Can not change number of buckets of a resilient nexthop group."); 2287 return -EINVAL; 2288 } 2289 2290 /* Emit a pre-replace notification so that listeners could veto 2291 * a potentially unsupported configuration. Otherwise, 2292 * individual bucket replacement notifications would need to be 2293 * vetoed, which is something that should only happen if the 2294 * bucket is currently active. 2295 */ 2296 err = call_nexthop_res_table_notifiers(net, new, extack); 2297 if (err) 2298 return err; 2299 2300 if (cfg->nh_grp_res_has_idle_timer) 2301 old_res_table->idle_timer = cfg->nh_grp_res_idle_timer; 2302 if (cfg->nh_grp_res_has_unbalanced_timer) 2303 old_res_table->unbalanced_timer = 2304 cfg->nh_grp_res_unbalanced_timer; 2305 2306 replace_nexthop_grp_res(oldg, newg); 2307 2308 tmp_table = new_res_table; 2309 rcu_assign_pointer(newg->res_table, old_res_table); 2310 rcu_assign_pointer(newg->spare->res_table, old_res_table); 2311 } 2312 2313 /* update parents - used by nexthop code for cleanup */ 2314 for (i = 0; i < newg->num_nh; i++) 2315 newg->nh_entries[i].nh_parent = old; 2316 2317 rcu_assign_pointer(old->nh_grp, newg); 2318 2319 /* Make sure concurrent readers are not using 'oldg' anymore. */ 2320 synchronize_net(); 2321 2322 if (newg->resilient) { 2323 rcu_assign_pointer(oldg->res_table, tmp_table); 2324 rcu_assign_pointer(oldg->spare->res_table, tmp_table); 2325 } 2326 2327 for (i = 0; i < oldg->num_nh; i++) 2328 oldg->nh_entries[i].nh_parent = new; 2329 2330 rcu_assign_pointer(new->nh_grp, oldg); 2331 2332 return 0; 2333} 2334 2335static void nh_group_v4_update(struct nh_group *nhg) 2336{ 2337 struct nh_grp_entry *nhges; 2338 bool has_v4 = false; 2339 int i; 2340 2341 nhges = nhg->nh_entries; 2342 for (i = 0; i < nhg->num_nh; i++) { 2343 struct nh_info *nhi; 2344 2345 nhi = rtnl_dereference(nhges[i].nh->nh_info); 2346 if (nhi->family == AF_INET) 2347 has_v4 = true; 2348 } 2349 nhg->has_v4 = has_v4; 2350} 2351 2352static int replace_nexthop_single_notify_res(struct net *net, 2353 struct nh_res_table *res_table, 2354 struct nexthop *old, 2355 struct nh_info *oldi, 2356 struct nh_info *newi, 2357 struct netlink_ext_ack *extack) 2358{ 2359 u32 nhg_id = res_table->nhg_id; 2360 int err; 2361 u16 i; 2362 2363 for (i = 0; i < res_table->num_nh_buckets; i++) { 2364 struct nh_res_bucket *bucket = &res_table->nh_buckets[i]; 2365 struct nh_grp_entry *nhge; 2366 2367 nhge = rtnl_dereference(bucket->nh_entry); 2368 if (nhge->nh == old) { 2369 err = __call_nexthop_res_bucket_notifiers(net, nhg_id, 2370 i, true, 2371 oldi, newi, 2372 extack); 2373 if (err) 2374 goto err_notify; 2375 } 2376 } 2377 2378 return 0; 2379 2380err_notify: 2381 while (i-- > 0) { 2382 struct nh_res_bucket *bucket = &res_table->nh_buckets[i]; 2383 struct nh_grp_entry *nhge; 2384 2385 nhge = rtnl_dereference(bucket->nh_entry); 2386 if (nhge->nh == old) 2387 __call_nexthop_res_bucket_notifiers(net, nhg_id, i, 2388 true, newi, oldi, 2389 extack); 2390 } 2391 return err; 2392} 2393 2394static int replace_nexthop_single_notify(struct net *net, 2395 struct nexthop *group_nh, 2396 struct nexthop *old, 2397 struct nh_info *oldi, 2398 struct nh_info *newi, 2399 struct netlink_ext_ack *extack) 2400{ 2401 struct nh_group *nhg = rtnl_dereference(group_nh->nh_grp); 2402 struct nh_res_table *res_table; 2403 2404 if (nhg->hash_threshold) { 2405 return call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, 2406 group_nh, extack); 2407 } else if (nhg->resilient) { 2408 res_table = rtnl_dereference(nhg->res_table); 2409 return replace_nexthop_single_notify_res(net, res_table, 2410 old, oldi, newi, 2411 extack); 2412 } 2413 2414 return -EINVAL; 2415} 2416 2417static int replace_nexthop_single(struct net *net, struct nexthop *old, 2418 struct nexthop *new, 2419 struct netlink_ext_ack *extack) 2420{ 2421 u8 old_protocol, old_nh_flags; 2422 struct nh_info *oldi, *newi; 2423 struct nh_grp_entry *nhge; 2424 int err; 2425 2426 if (new->is_group) { 2427 NL_SET_ERR_MSG(extack, "Can not replace a nexthop with a nexthop group."); 2428 return -EINVAL; 2429 } 2430 2431 if (!list_empty(&old->grp_list) && 2432 rtnl_dereference(new->nh_info)->fdb_nh != 2433 rtnl_dereference(old->nh_info)->fdb_nh) { 2434 NL_SET_ERR_MSG(extack, "Cannot change nexthop FDB status while in a group"); 2435 return -EINVAL; 2436 } 2437 2438 err = call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, new, extack); 2439 if (err) 2440 return err; 2441 2442 /* Hardware flags were set on 'old' as 'new' is not in the red-black 2443 * tree. Therefore, inherit the flags from 'old' to 'new'. 2444 */ 2445 new->nh_flags |= old->nh_flags & (RTNH_F_OFFLOAD | RTNH_F_TRAP); 2446 2447 oldi = rtnl_dereference(old->nh_info); 2448 newi = rtnl_dereference(new->nh_info); 2449 2450 newi->nh_parent = old; 2451 oldi->nh_parent = new; 2452 2453 old_protocol = old->protocol; 2454 old_nh_flags = old->nh_flags; 2455 2456 old->protocol = new->protocol; 2457 old->nh_flags = new->nh_flags; 2458 2459 rcu_assign_pointer(old->nh_info, newi); 2460 rcu_assign_pointer(new->nh_info, oldi); 2461 2462 /* Send a replace notification for all the groups using the nexthop. */ 2463 list_for_each_entry(nhge, &old->grp_list, nh_list) { 2464 struct nexthop *nhp = nhge->nh_parent; 2465 2466 err = replace_nexthop_single_notify(net, nhp, old, oldi, newi, 2467 extack); 2468 if (err) 2469 goto err_notify; 2470 } 2471 2472 /* When replacing an IPv4 nexthop with an IPv6 nexthop, potentially 2473 * update IPv4 indication in all the groups using the nexthop. 2474 */ 2475 if (oldi->family == AF_INET && newi->family == AF_INET6) { 2476 list_for_each_entry(nhge, &old->grp_list, nh_list) { 2477 struct nexthop *nhp = nhge->nh_parent; 2478 struct nh_group *nhg; 2479 2480 nhg = rtnl_dereference(nhp->nh_grp); 2481 nh_group_v4_update(nhg); 2482 } 2483 } 2484 2485 return 0; 2486 2487err_notify: 2488 rcu_assign_pointer(new->nh_info, newi); 2489 rcu_assign_pointer(old->nh_info, oldi); 2490 old->nh_flags = old_nh_flags; 2491 old->protocol = old_protocol; 2492 oldi->nh_parent = old; 2493 newi->nh_parent = new; 2494 list_for_each_entry_continue_reverse(nhge, &old->grp_list, nh_list) { 2495 struct nexthop *nhp = nhge->nh_parent; 2496 2497 replace_nexthop_single_notify(net, nhp, old, newi, oldi, NULL); 2498 } 2499 call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, old, extack); 2500 return err; 2501} 2502 2503static void __nexthop_replace_notify(struct net *net, struct nexthop *nh, 2504 struct nl_info *info) 2505{ 2506 struct fib6_info *f6i; 2507 2508 if (!list_empty(&nh->fi_list)) { 2509 struct fib_info *fi; 2510 2511 /* expectation is a few fib_info per nexthop and then 2512 * a lot of routes per fib_info. So mark the fib_info 2513 * and then walk the fib tables once 2514 */ 2515 list_for_each_entry(fi, &nh->fi_list, nh_list) 2516 fi->nh_updated = true; 2517 2518 fib_info_notify_update(net, info); 2519 2520 list_for_each_entry(fi, &nh->fi_list, nh_list) 2521 fi->nh_updated = false; 2522 } 2523 2524 list_for_each_entry(f6i, &nh->f6i_list, nh_list) 2525 ipv6_stub->fib6_rt_update(net, f6i, info); 2526} 2527 2528/* send RTM_NEWROUTE with REPLACE flag set for all FIB entries 2529 * linked to this nexthop and for all groups that the nexthop 2530 * is a member of 2531 */ 2532static void nexthop_replace_notify(struct net *net, struct nexthop *nh, 2533 struct nl_info *info) 2534{ 2535 struct nh_grp_entry *nhge; 2536 2537 __nexthop_replace_notify(net, nh, info); 2538 2539 list_for_each_entry(nhge, &nh->grp_list, nh_list) 2540 __nexthop_replace_notify(net, nhge->nh_parent, info); 2541} 2542 2543static int replace_nexthop(struct net *net, struct nexthop *old, 2544 struct nexthop *new, const struct nh_config *cfg, 2545 struct netlink_ext_ack *extack) 2546{ 2547 bool new_is_reject = false; 2548 struct nh_grp_entry *nhge; 2549 int err; 2550 2551 /* check that existing FIB entries are ok with the 2552 * new nexthop definition 2553 */ 2554 err = fib_check_nh_list(old, new, extack); 2555 if (err) 2556 return err; 2557 2558 err = fib6_check_nh_list(old, new, extack); 2559 if (err) 2560 return err; 2561 2562 if (!new->is_group) { 2563 struct nh_info *nhi = rtnl_dereference(new->nh_info); 2564 2565 new_is_reject = nhi->reject_nh; 2566 } 2567 2568 list_for_each_entry(nhge, &old->grp_list, nh_list) { 2569 /* if new nexthop is a blackhole, any groups using this 2570 * nexthop cannot have more than 1 path 2571 */ 2572 if (new_is_reject && 2573 nexthop_num_path(nhge->nh_parent) > 1) { 2574 NL_SET_ERR_MSG(extack, "Blackhole nexthop can not be a member of a group with more than one path"); 2575 return -EINVAL; 2576 } 2577 2578 err = fib_check_nh_list(nhge->nh_parent, new, extack); 2579 if (err) 2580 return err; 2581 2582 err = fib6_check_nh_list(nhge->nh_parent, new, extack); 2583 if (err) 2584 return err; 2585 } 2586 2587 if (old->is_group) 2588 err = replace_nexthop_grp(net, old, new, cfg, extack); 2589 else 2590 err = replace_nexthop_single(net, old, new, extack); 2591 2592 if (!err) { 2593 nh_rt_cache_flush(net, old, new); 2594 2595 __remove_nexthop(net, new, NULL); 2596 nexthop_put(new); 2597 } 2598 2599 return err; 2600} 2601 2602/* called with rtnl_lock held */ 2603static int insert_nexthop(struct net *net, struct nexthop *new_nh, 2604 struct nh_config *cfg, struct netlink_ext_ack *extack) 2605{ 2606 struct rb_node **pp, *parent = NULL, *next; 2607 struct rb_root *root = &net->nexthop.rb_root; 2608 bool replace = !!(cfg->nlflags & NLM_F_REPLACE); 2609 bool create = !!(cfg->nlflags & NLM_F_CREATE); 2610 u32 new_id = new_nh->id; 2611 int replace_notify = 0; 2612 int rc = -EEXIST; 2613 2614 pp = &root->rb_node; 2615 while (1) { 2616 struct nexthop *nh; 2617 2618 next = *pp; 2619 if (!next) 2620 break; 2621 2622 parent = next; 2623 2624 nh = rb_entry(parent, struct nexthop, rb_node); 2625 if (new_id < nh->id) { 2626 pp = &next->rb_left; 2627 } else if (new_id > nh->id) { 2628 pp = &next->rb_right; 2629 } else if (replace) { 2630 rc = replace_nexthop(net, nh, new_nh, cfg, extack); 2631 if (!rc) { 2632 new_nh = nh; /* send notification with old nh */ 2633 replace_notify = 1; 2634 } 2635 goto out; 2636 } else { 2637 /* id already exists and not a replace */ 2638 goto out; 2639 } 2640 } 2641 2642 if (replace && !create) { 2643 NL_SET_ERR_MSG(extack, "Replace specified without create and no entry exists"); 2644 rc = -ENOENT; 2645 goto out; 2646 } 2647 2648 if (new_nh->is_group) { 2649 struct nh_group *nhg = rtnl_dereference(new_nh->nh_grp); 2650 struct nh_res_table *res_table; 2651 2652 if (nhg->resilient) { 2653 res_table = rtnl_dereference(nhg->res_table); 2654 2655 /* Not passing the number of buckets is OK when 2656 * replacing, but not when creating a new group. 2657 */ 2658 if (!cfg->nh_grp_res_has_num_buckets) { 2659 NL_SET_ERR_MSG(extack, "Number of buckets not specified for nexthop group insertion"); 2660 rc = -EINVAL; 2661 goto out; 2662 } 2663 2664 nh_res_group_rebalance(nhg, res_table); 2665 2666 /* Do not send bucket notifications, we do full 2667 * notification below. 2668 */ 2669 nh_res_table_upkeep(res_table, false, false); 2670 } 2671 } 2672 2673 rb_link_node_rcu(&new_nh->rb_node, parent, pp); 2674 rb_insert_color(&new_nh->rb_node, root); 2675 2676 /* The initial insertion is a full notification for hash-threshold as 2677 * well as resilient groups. 2678 */ 2679 rc = call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, new_nh, extack); 2680 if (rc) 2681 rb_erase(&new_nh->rb_node, &net->nexthop.rb_root); 2682 2683out: 2684 if (!rc) { 2685 nh_base_seq_inc(net); 2686 nexthop_notify(RTM_NEWNEXTHOP, new_nh, &cfg->nlinfo); 2687 if (replace_notify && 2688 READ_ONCE(net->ipv4.sysctl_nexthop_compat_mode)) 2689 nexthop_replace_notify(net, new_nh, &cfg->nlinfo); 2690 } 2691 2692 return rc; 2693} 2694 2695/* rtnl */ 2696/* remove all nexthops tied to a device being deleted */ 2697static void nexthop_flush_dev(struct net_device *dev, unsigned long event) 2698{ 2699 unsigned int hash = nh_dev_hashfn(dev->ifindex); 2700 struct net *net = dev_net(dev); 2701 struct hlist_head *head = &net->nexthop.devhash[hash]; 2702 struct hlist_node *n; 2703 struct nh_info *nhi; 2704 2705 hlist_for_each_entry_safe(nhi, n, head, dev_hash) { 2706 if (nhi->fib_nhc.nhc_dev != dev) 2707 continue; 2708 2709 if (nhi->reject_nh && 2710 (event == NETDEV_DOWN || event == NETDEV_CHANGE)) 2711 continue; 2712 2713 remove_nexthop(net, nhi->nh_parent, NULL); 2714 } 2715} 2716 2717/* rtnl; called when net namespace is deleted */ 2718static void flush_all_nexthops(struct net *net) 2719{ 2720 struct rb_root *root = &net->nexthop.rb_root; 2721 struct rb_node *node; 2722 struct nexthop *nh; 2723 2724 while ((node = rb_first(root))) { 2725 nh = rb_entry(node, struct nexthop, rb_node); 2726 remove_nexthop(net, nh, NULL); 2727 cond_resched(); 2728 } 2729} 2730 2731static struct nexthop *nexthop_create_group(struct net *net, 2732 struct nh_config *cfg) 2733{ 2734 struct nlattr *grps_attr = cfg->nh_grp; 2735 struct nexthop_grp *entry = nla_data(grps_attr); 2736 u16 num_nh = nla_len(grps_attr) / sizeof(*entry); 2737 struct nh_group *nhg; 2738 struct nexthop *nh; 2739 int err; 2740 int i; 2741 2742 nh = nexthop_alloc(); 2743 if (!nh) 2744 return ERR_PTR(-ENOMEM); 2745 2746 nh->is_group = 1; 2747 2748 nhg = nexthop_grp_alloc(num_nh); 2749 if (!nhg) { 2750 kfree(nh); 2751 return ERR_PTR(-ENOMEM); 2752 } 2753 2754 /* spare group used for removals */ 2755 nhg->spare = nexthop_grp_alloc(num_nh); 2756 if (!nhg->spare) { 2757 kfree(nhg); 2758 kfree(nh); 2759 return ERR_PTR(-ENOMEM); 2760 } 2761 nhg->spare->spare = nhg; 2762 2763 for (i = 0; i < nhg->num_nh; ++i) { 2764 struct nexthop *nhe; 2765 struct nh_info *nhi; 2766 2767 nhe = nexthop_find_by_id(net, entry[i].id); 2768 if (!nexthop_get(nhe)) { 2769 err = -ENOENT; 2770 goto out_no_nh; 2771 } 2772 2773 nhi = rtnl_dereference(nhe->nh_info); 2774 if (nhi->family == AF_INET) 2775 nhg->has_v4 = true; 2776 2777 nhg->nh_entries[i].stats = 2778 netdev_alloc_pcpu_stats(struct nh_grp_entry_stats); 2779 if (!nhg->nh_entries[i].stats) { 2780 err = -ENOMEM; 2781 nexthop_put(nhe); 2782 goto out_no_nh; 2783 } 2784 nhg->nh_entries[i].nh = nhe; 2785 nhg->nh_entries[i].weight = nexthop_grp_weight(&entry[i]); 2786 2787 list_add(&nhg->nh_entries[i].nh_list, &nhe->grp_list); 2788 nhg->nh_entries[i].nh_parent = nh; 2789 } 2790 2791 if (cfg->nh_grp_type == NEXTHOP_GRP_TYPE_MPATH) { 2792 nhg->hash_threshold = 1; 2793 nhg->is_multipath = true; 2794 } else if (cfg->nh_grp_type == NEXTHOP_GRP_TYPE_RES) { 2795 struct nh_res_table *res_table; 2796 2797 res_table = nexthop_res_table_alloc(net, cfg->nh_id, cfg); 2798 if (!res_table) { 2799 err = -ENOMEM; 2800 goto out_no_nh; 2801 } 2802 2803 rcu_assign_pointer(nhg->spare->res_table, res_table); 2804 rcu_assign_pointer(nhg->res_table, res_table); 2805 nhg->resilient = true; 2806 nhg->is_multipath = true; 2807 } 2808 2809 WARN_ON_ONCE(nhg->hash_threshold + nhg->resilient != 1); 2810 2811 if (nhg->hash_threshold) 2812 nh_hthr_group_rebalance(nhg); 2813 2814 if (cfg->nh_fdb) 2815 nhg->fdb_nh = 1; 2816 2817 if (cfg->nh_hw_stats) 2818 nhg->hw_stats = true; 2819 2820 rcu_assign_pointer(nh->nh_grp, nhg); 2821 2822 return nh; 2823 2824out_no_nh: 2825 for (i--; i >= 0; --i) { 2826 list_del(&nhg->nh_entries[i].nh_list); 2827 free_percpu(nhg->nh_entries[i].stats); 2828 nexthop_put(nhg->nh_entries[i].nh); 2829 } 2830 2831 kfree(nhg->spare); 2832 kfree(nhg); 2833 kfree(nh); 2834 2835 return ERR_PTR(err); 2836} 2837 2838static int nh_create_ipv4(struct net *net, struct nexthop *nh, 2839 struct nh_info *nhi, struct nh_config *cfg, 2840 struct netlink_ext_ack *extack) 2841{ 2842 struct fib_nh *fib_nh = &nhi->fib_nh; 2843 struct fib_config fib_cfg = { 2844 .fc_oif = cfg->nh_ifindex, 2845 .fc_gw4 = cfg->gw.ipv4, 2846 .fc_gw_family = cfg->gw.ipv4 ? AF_INET : 0, 2847 .fc_flags = cfg->nh_flags, 2848 .fc_nlinfo = cfg->nlinfo, 2849 .fc_encap = cfg->nh_encap, 2850 .fc_encap_type = cfg->nh_encap_type, 2851 }; 2852 u32 tb_id = (cfg->dev ? l3mdev_fib_table(cfg->dev) : RT_TABLE_MAIN); 2853 int err; 2854 2855 err = fib_nh_init(net, fib_nh, &fib_cfg, 1, extack); 2856 if (err) { 2857 fib_nh_release(net, fib_nh); 2858 goto out; 2859 } 2860 2861 if (nhi->fdb_nh) 2862 goto out; 2863 2864 /* sets nh_dev if successful */ 2865 err = fib_check_nh(net, fib_nh, tb_id, 0, extack); 2866 if (!err) { 2867 nh->nh_flags = fib_nh->fib_nh_flags; 2868 fib_info_update_nhc_saddr(net, &fib_nh->nh_common, 2869 !fib_nh->fib_nh_scope ? 0 : fib_nh->fib_nh_scope - 1); 2870 } else { 2871 fib_nh_release(net, fib_nh); 2872 } 2873out: 2874 return err; 2875} 2876 2877static int nh_create_ipv6(struct net *net, struct nexthop *nh, 2878 struct nh_info *nhi, struct nh_config *cfg, 2879 struct netlink_ext_ack *extack) 2880{ 2881 struct fib6_nh *fib6_nh = &nhi->fib6_nh; 2882 struct fib6_config fib6_cfg = { 2883 .fc_table = l3mdev_fib_table(cfg->dev), 2884 .fc_ifindex = cfg->nh_ifindex, 2885 .fc_gateway = cfg->gw.ipv6, 2886 .fc_flags = cfg->nh_flags, 2887 .fc_nlinfo = cfg->nlinfo, 2888 .fc_encap = cfg->nh_encap, 2889 .fc_encap_type = cfg->nh_encap_type, 2890 .fc_is_fdb = cfg->nh_fdb, 2891 }; 2892 int err; 2893 2894 if (!ipv6_addr_any(&cfg->gw.ipv6)) 2895 fib6_cfg.fc_flags |= RTF_GATEWAY; 2896 2897 /* sets nh_dev if successful */ 2898 err = ipv6_stub->fib6_nh_init(net, fib6_nh, &fib6_cfg, GFP_KERNEL, 2899 extack); 2900 if (err) { 2901 /* IPv6 is not enabled, don't call fib6_nh_release */ 2902 if (err == -EAFNOSUPPORT) 2903 goto out; 2904 ipv6_stub->fib6_nh_release(fib6_nh); 2905 } else { 2906 nh->nh_flags = fib6_nh->fib_nh_flags; 2907 } 2908out: 2909 return err; 2910} 2911 2912static struct nexthop *nexthop_create(struct net *net, struct nh_config *cfg, 2913 struct netlink_ext_ack *extack) 2914{ 2915 struct nh_info *nhi; 2916 struct nexthop *nh; 2917 int err = 0; 2918 2919 nh = nexthop_alloc(); 2920 if (!nh) 2921 return ERR_PTR(-ENOMEM); 2922 2923 nhi = kzalloc(sizeof(*nhi), GFP_KERNEL); 2924 if (!nhi) { 2925 kfree(nh); 2926 return ERR_PTR(-ENOMEM); 2927 } 2928 2929 nh->nh_flags = cfg->nh_flags; 2930 nh->net = net; 2931 2932 nhi->nh_parent = nh; 2933 nhi->family = cfg->nh_family; 2934 nhi->fib_nhc.nhc_scope = RT_SCOPE_LINK; 2935 2936 if (cfg->nh_fdb) 2937 nhi->fdb_nh = 1; 2938 2939 if (cfg->nh_blackhole) { 2940 nhi->reject_nh = 1; 2941 cfg->nh_ifindex = net->loopback_dev->ifindex; 2942 } 2943 2944 switch (cfg->nh_family) { 2945 case AF_INET: 2946 err = nh_create_ipv4(net, nh, nhi, cfg, extack); 2947 break; 2948 case AF_INET6: 2949 err = nh_create_ipv6(net, nh, nhi, cfg, extack); 2950 break; 2951 } 2952 2953 if (err) { 2954 kfree(nhi); 2955 kfree(nh); 2956 return ERR_PTR(err); 2957 } 2958 2959 /* add the entry to the device based hash */ 2960 if (!nhi->fdb_nh) 2961 nexthop_devhash_add(net, nhi); 2962 2963 rcu_assign_pointer(nh->nh_info, nhi); 2964 2965 return nh; 2966} 2967 2968/* called with rtnl lock held */ 2969static struct nexthop *nexthop_add(struct net *net, struct nh_config *cfg, 2970 struct netlink_ext_ack *extack) 2971{ 2972 struct nexthop *nh; 2973 int err; 2974 2975 if (!cfg->nh_id) { 2976 cfg->nh_id = nh_find_unused_id(net); 2977 if (!cfg->nh_id) { 2978 NL_SET_ERR_MSG(extack, "No unused id"); 2979 return ERR_PTR(-EINVAL); 2980 } 2981 } 2982 2983 if (cfg->nh_grp) 2984 nh = nexthop_create_group(net, cfg); 2985 else 2986 nh = nexthop_create(net, cfg, extack); 2987 2988 if (IS_ERR(nh)) 2989 return nh; 2990 2991 refcount_set(&nh->refcnt, 1); 2992 nh->id = cfg->nh_id; 2993 nh->protocol = cfg->nh_protocol; 2994 nh->net = net; 2995 2996 err = insert_nexthop(net, nh, cfg, extack); 2997 if (err) { 2998 __remove_nexthop(net, nh, NULL); 2999 nexthop_put(nh); 3000 nh = ERR_PTR(err); 3001 } 3002 3003 return nh; 3004} 3005 3006static int rtm_nh_get_timer(struct nlattr *attr, unsigned long fallback, 3007 unsigned long *timer_p, bool *has_p, 3008 struct netlink_ext_ack *extack) 3009{ 3010 unsigned long timer; 3011 u32 value; 3012 3013 if (!attr) { 3014 *timer_p = fallback; 3015 *has_p = false; 3016 return 0; 3017 } 3018 3019 value = nla_get_u32(attr); 3020 timer = clock_t_to_jiffies(value); 3021 if (timer == ~0UL) { 3022 NL_SET_ERR_MSG(extack, "Timer value too large"); 3023 return -EINVAL; 3024 } 3025 3026 *timer_p = timer; 3027 *has_p = true; 3028 return 0; 3029} 3030 3031static int rtm_to_nh_config_grp_res(struct nlattr *res, struct nh_config *cfg, 3032 struct netlink_ext_ack *extack) 3033{ 3034 struct nlattr *tb[ARRAY_SIZE(rtm_nh_res_policy_new)] = {}; 3035 int err; 3036 3037 if (res) { 3038 err = nla_parse_nested(tb, 3039 ARRAY_SIZE(rtm_nh_res_policy_new) - 1, 3040 res, rtm_nh_res_policy_new, extack); 3041 if (err < 0) 3042 return err; 3043 } 3044 3045 if (tb[NHA_RES_GROUP_BUCKETS]) { 3046 cfg->nh_grp_res_num_buckets = 3047 nla_get_u16(tb[NHA_RES_GROUP_BUCKETS]); 3048 cfg->nh_grp_res_has_num_buckets = true; 3049 if (!cfg->nh_grp_res_num_buckets) { 3050 NL_SET_ERR_MSG(extack, "Number of buckets needs to be non-0"); 3051 return -EINVAL; 3052 } 3053 } 3054 3055 err = rtm_nh_get_timer(tb[NHA_RES_GROUP_IDLE_TIMER], 3056 NH_RES_DEFAULT_IDLE_TIMER, 3057 &cfg->nh_grp_res_idle_timer, 3058 &cfg->nh_grp_res_has_idle_timer, 3059 extack); 3060 if (err) 3061 return err; 3062 3063 return rtm_nh_get_timer(tb[NHA_RES_GROUP_UNBALANCED_TIMER], 3064 NH_RES_DEFAULT_UNBALANCED_TIMER, 3065 &cfg->nh_grp_res_unbalanced_timer, 3066 &cfg->nh_grp_res_has_unbalanced_timer, 3067 extack); 3068} 3069 3070static int rtm_to_nh_config(struct net *net, struct sk_buff *skb, 3071 struct nlmsghdr *nlh, struct nlattr **tb, 3072 struct nh_config *cfg, 3073 struct netlink_ext_ack *extack) 3074{ 3075 struct nhmsg *nhm = nlmsg_data(nlh); 3076 int err; 3077 3078 err = -EINVAL; 3079 if (nhm->resvd || nhm->nh_scope) { 3080 NL_SET_ERR_MSG(extack, "Invalid values in ancillary header"); 3081 goto out; 3082 } 3083 if (nhm->nh_flags & ~NEXTHOP_VALID_USER_FLAGS) { 3084 NL_SET_ERR_MSG(extack, "Invalid nexthop flags in ancillary header"); 3085 goto out; 3086 } 3087 3088 switch (nhm->nh_family) { 3089 case AF_INET: 3090 case AF_INET6: 3091 break; 3092 case AF_UNSPEC: 3093 if (tb[NHA_GROUP]) 3094 break; 3095 fallthrough; 3096 default: 3097 NL_SET_ERR_MSG(extack, "Invalid address family"); 3098 goto out; 3099 } 3100 3101 memset(cfg, 0, sizeof(*cfg)); 3102 cfg->nlflags = nlh->nlmsg_flags; 3103 cfg->nlinfo.portid = NETLINK_CB(skb).portid; 3104 cfg->nlinfo.nlh = nlh; 3105 cfg->nlinfo.nl_net = net; 3106 3107 cfg->nh_family = nhm->nh_family; 3108 cfg->nh_protocol = nhm->nh_protocol; 3109 cfg->nh_flags = nhm->nh_flags; 3110 3111 if (tb[NHA_ID]) 3112 cfg->nh_id = nla_get_u32(tb[NHA_ID]); 3113 3114 if (tb[NHA_FDB]) { 3115 if (tb[NHA_OIF] || tb[NHA_BLACKHOLE] || 3116 tb[NHA_ENCAP] || tb[NHA_ENCAP_TYPE]) { 3117 NL_SET_ERR_MSG(extack, "Fdb attribute can not be used with encap, oif or blackhole"); 3118 goto out; 3119 } 3120 if (nhm->nh_flags) { 3121 NL_SET_ERR_MSG(extack, "Unsupported nexthop flags in ancillary header"); 3122 goto out; 3123 } 3124 cfg->nh_fdb = nla_get_flag(tb[NHA_FDB]); 3125 } 3126 3127 if (tb[NHA_GROUP]) { 3128 if (nhm->nh_family != AF_UNSPEC) { 3129 NL_SET_ERR_MSG(extack, "Invalid family for group"); 3130 goto out; 3131 } 3132 cfg->nh_grp = tb[NHA_GROUP]; 3133 3134 cfg->nh_grp_type = NEXTHOP_GRP_TYPE_MPATH; 3135 if (tb[NHA_GROUP_TYPE]) 3136 cfg->nh_grp_type = nla_get_u16(tb[NHA_GROUP_TYPE]); 3137 3138 if (cfg->nh_grp_type > NEXTHOP_GRP_TYPE_MAX) { 3139 NL_SET_ERR_MSG(extack, "Invalid group type"); 3140 goto out; 3141 } 3142 3143 err = nh_check_attr_group(net, tb, ARRAY_SIZE(rtm_nh_policy_new), 3144 cfg->nh_grp_type, extack); 3145 if (err) 3146 goto out; 3147 3148 if (cfg->nh_grp_type == NEXTHOP_GRP_TYPE_RES) 3149 err = rtm_to_nh_config_grp_res(tb[NHA_RES_GROUP], 3150 cfg, extack); 3151 3152 if (tb[NHA_HW_STATS_ENABLE]) 3153 cfg->nh_hw_stats = nla_get_u32(tb[NHA_HW_STATS_ENABLE]); 3154 3155 /* no other attributes should be set */ 3156 goto out; 3157 } 3158 3159 if (tb[NHA_BLACKHOLE]) { 3160 if (tb[NHA_GATEWAY] || tb[NHA_OIF] || 3161 tb[NHA_ENCAP] || tb[NHA_ENCAP_TYPE] || tb[NHA_FDB]) { 3162 NL_SET_ERR_MSG(extack, "Blackhole attribute can not be used with gateway, oif, encap or fdb"); 3163 goto out; 3164 } 3165 3166 cfg->nh_blackhole = 1; 3167 err = 0; 3168 goto out; 3169 } 3170 3171 if (!cfg->nh_fdb && !tb[NHA_OIF]) { 3172 NL_SET_ERR_MSG(extack, "Device attribute required for non-blackhole and non-fdb nexthops"); 3173 goto out; 3174 } 3175 3176 err = -EINVAL; 3177 if (tb[NHA_GATEWAY]) { 3178 struct nlattr *gwa = tb[NHA_GATEWAY]; 3179 3180 switch (cfg->nh_family) { 3181 case AF_INET: 3182 if (nla_len(gwa) != sizeof(u32)) { 3183 NL_SET_ERR_MSG(extack, "Invalid gateway"); 3184 goto out; 3185 } 3186 cfg->gw.ipv4 = nla_get_be32(gwa); 3187 break; 3188 case AF_INET6: 3189 if (nla_len(gwa) != sizeof(struct in6_addr)) { 3190 NL_SET_ERR_MSG(extack, "Invalid gateway"); 3191 goto out; 3192 } 3193 cfg->gw.ipv6 = nla_get_in6_addr(gwa); 3194 break; 3195 default: 3196 NL_SET_ERR_MSG(extack, 3197 "Unknown address family for gateway"); 3198 goto out; 3199 } 3200 } else { 3201 /* device only nexthop (no gateway) */ 3202 if (cfg->nh_flags & RTNH_F_ONLINK) { 3203 NL_SET_ERR_MSG(extack, 3204 "ONLINK flag can not be set for nexthop without a gateway"); 3205 goto out; 3206 } 3207 } 3208 3209 if (tb[NHA_ENCAP]) { 3210 cfg->nh_encap = tb[NHA_ENCAP]; 3211 3212 if (!tb[NHA_ENCAP_TYPE]) { 3213 NL_SET_ERR_MSG(extack, "LWT encapsulation type is missing"); 3214 goto out; 3215 } 3216 3217 cfg->nh_encap_type = nla_get_u16(tb[NHA_ENCAP_TYPE]); 3218 err = lwtunnel_valid_encap_type(cfg->nh_encap_type, extack); 3219 if (err < 0) 3220 goto out; 3221 3222 } else if (tb[NHA_ENCAP_TYPE]) { 3223 NL_SET_ERR_MSG(extack, "LWT encapsulation attribute is missing"); 3224 goto out; 3225 } 3226 3227 if (tb[NHA_HW_STATS_ENABLE]) { 3228 NL_SET_ERR_MSG(extack, "Cannot enable nexthop hardware statistics for non-group nexthops"); 3229 goto out; 3230 } 3231 3232 err = 0; 3233out: 3234 return err; 3235} 3236 3237static int rtm_to_nh_config_rtnl(struct net *net, struct nlattr **tb, 3238 struct nh_config *cfg, 3239 struct netlink_ext_ack *extack) 3240{ 3241 if (tb[NHA_GROUP]) 3242 return nh_check_attr_group_rtnl(net, tb, extack); 3243 3244 if (tb[NHA_OIF]) { 3245 cfg->nh_ifindex = nla_get_u32(tb[NHA_OIF]); 3246 if (cfg->nh_ifindex) 3247 cfg->dev = __dev_get_by_index(net, cfg->nh_ifindex); 3248 3249 if (!cfg->dev) { 3250 NL_SET_ERR_MSG(extack, "Invalid device index"); 3251 return -EINVAL; 3252 } 3253 3254 if (!(cfg->dev->flags & IFF_UP)) { 3255 NL_SET_ERR_MSG(extack, "Nexthop device is not up"); 3256 return -ENETDOWN; 3257 } 3258 3259 if (!netif_carrier_ok(cfg->dev)) { 3260 NL_SET_ERR_MSG(extack, "Carrier for nexthop device is down"); 3261 return -ENETDOWN; 3262 } 3263 } 3264 3265 return 0; 3266} 3267 3268/* rtnl */ 3269static int rtm_new_nexthop(struct sk_buff *skb, struct nlmsghdr *nlh, 3270 struct netlink_ext_ack *extack) 3271{ 3272 struct nlattr *tb[ARRAY_SIZE(rtm_nh_policy_new)]; 3273 struct net *net = sock_net(skb->sk); 3274 struct nh_config cfg; 3275 struct nexthop *nh; 3276 int err; 3277 3278 err = nlmsg_parse(nlh, sizeof(struct nhmsg), tb, 3279 ARRAY_SIZE(rtm_nh_policy_new) - 1, 3280 rtm_nh_policy_new, extack); 3281 if (err < 0) 3282 goto out; 3283 3284 err = rtm_to_nh_config(net, skb, nlh, tb, &cfg, extack); 3285 if (err) 3286 goto out; 3287 3288 if (cfg.nlflags & NLM_F_REPLACE && !cfg.nh_id) { 3289 NL_SET_ERR_MSG(extack, "Replace requires nexthop id"); 3290 err = -EINVAL; 3291 goto out; 3292 } 3293 3294 rtnl_net_lock(net); 3295 3296 err = rtm_to_nh_config_rtnl(net, tb, &cfg, extack); 3297 if (err) 3298 goto unlock; 3299 3300 nh = nexthop_add(net, &cfg, extack); 3301 if (IS_ERR(nh)) 3302 err = PTR_ERR(nh); 3303 3304unlock: 3305 rtnl_net_unlock(net); 3306out: 3307 return err; 3308} 3309 3310static int nh_valid_get_del_req(const struct nlmsghdr *nlh, 3311 struct nlattr **tb, u32 *id, u32 *op_flags, 3312 struct netlink_ext_ack *extack) 3313{ 3314 struct nhmsg *nhm = nlmsg_data(nlh); 3315 3316 if (nhm->nh_protocol || nhm->resvd || nhm->nh_scope || nhm->nh_flags) { 3317 NL_SET_ERR_MSG(extack, "Invalid values in header"); 3318 return -EINVAL; 3319 } 3320 3321 if (!tb[NHA_ID]) { 3322 NL_SET_ERR_MSG(extack, "Nexthop id is missing"); 3323 return -EINVAL; 3324 } 3325 3326 *id = nla_get_u32(tb[NHA_ID]); 3327 if (!(*id)) { 3328 NL_SET_ERR_MSG(extack, "Invalid nexthop id"); 3329 return -EINVAL; 3330 } 3331 3332 if (op_flags) 3333 *op_flags = nla_get_u32_default(tb[NHA_OP_FLAGS], 0); 3334 3335 return 0; 3336} 3337 3338/* rtnl */ 3339static int rtm_del_nexthop(struct sk_buff *skb, struct nlmsghdr *nlh, 3340 struct netlink_ext_ack *extack) 3341{ 3342 struct nlattr *tb[ARRAY_SIZE(rtm_nh_policy_del)]; 3343 struct net *net = sock_net(skb->sk); 3344 struct nl_info nlinfo = { 3345 .nlh = nlh, 3346 .nl_net = net, 3347 .portid = NETLINK_CB(skb).portid, 3348 }; 3349 struct nexthop *nh; 3350 int err; 3351 u32 id; 3352 3353 err = nlmsg_parse(nlh, sizeof(struct nhmsg), tb, 3354 ARRAY_SIZE(rtm_nh_policy_del) - 1, rtm_nh_policy_del, 3355 extack); 3356 if (err < 0) 3357 return err; 3358 3359 err = nh_valid_get_del_req(nlh, tb, &id, NULL, extack); 3360 if (err) 3361 return err; 3362 3363 rtnl_net_lock(net); 3364 3365 nh = nexthop_find_by_id(net, id); 3366 if (nh) 3367 remove_nexthop(net, nh, &nlinfo); 3368 else 3369 err = -ENOENT; 3370 3371 rtnl_net_unlock(net); 3372 3373 return err; 3374} 3375 3376/* rtnl */ 3377static int rtm_get_nexthop(struct sk_buff *in_skb, struct nlmsghdr *nlh, 3378 struct netlink_ext_ack *extack) 3379{ 3380 struct nlattr *tb[ARRAY_SIZE(rtm_nh_policy_get)]; 3381 struct net *net = sock_net(in_skb->sk); 3382 struct sk_buff *skb = NULL; 3383 struct nexthop *nh; 3384 u32 op_flags; 3385 int err; 3386 u32 id; 3387 3388 err = nlmsg_parse(nlh, sizeof(struct nhmsg), tb, 3389 ARRAY_SIZE(rtm_nh_policy_get) - 1, rtm_nh_policy_get, 3390 extack); 3391 if (err < 0) 3392 return err; 3393 3394 err = nh_valid_get_del_req(nlh, tb, &id, &op_flags, extack); 3395 if (err) 3396 return err; 3397 3398 err = -ENOENT; 3399 nh = nexthop_find_by_id(net, id); 3400 if (!nh) 3401 goto out; 3402 3403 err = -ENOBUFS; 3404 skb = nlmsg_new(nh_nlmsg_size(nh, op_flags), GFP_KERNEL); 3405 if (!skb) 3406 goto out; 3407 3408 err = nh_fill_node(skb, nh, RTM_NEWNEXTHOP, NETLINK_CB(in_skb).portid, 3409 nlh->nlmsg_seq, 0, op_flags); 3410 if (err < 0) { 3411 WARN_ON(err == -EMSGSIZE); 3412 goto errout_free; 3413 } 3414 3415 err = rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid); 3416out: 3417 return err; 3418errout_free: 3419 kfree_skb(skb); 3420 goto out; 3421} 3422 3423struct nh_dump_filter { 3424 u32 nh_id; 3425 int dev_idx; 3426 int master_idx; 3427 bool group_filter; 3428 bool fdb_filter; 3429 u32 res_bucket_nh_id; 3430 u32 op_flags; 3431}; 3432 3433static bool nh_dump_filtered(struct nexthop *nh, 3434 struct nh_dump_filter *filter, u8 family) 3435{ 3436 const struct net_device *dev; 3437 const struct nh_info *nhi; 3438 3439 if (filter->group_filter && !nh->is_group) 3440 return true; 3441 3442 if (!filter->dev_idx && !filter->master_idx && !family) 3443 return false; 3444 3445 if (nh->is_group) 3446 return true; 3447 3448 nhi = rtnl_dereference(nh->nh_info); 3449 if (family && nhi->family != family) 3450 return true; 3451 3452 dev = nhi->fib_nhc.nhc_dev; 3453 if (filter->dev_idx && (!dev || dev->ifindex != filter->dev_idx)) 3454 return true; 3455 3456 if (filter->master_idx) { 3457 struct net_device *master; 3458 3459 if (!dev) 3460 return true; 3461 3462 master = netdev_master_upper_dev_get((struct net_device *)dev); 3463 if (!master || master->ifindex != filter->master_idx) 3464 return true; 3465 } 3466 3467 return false; 3468} 3469 3470static int __nh_valid_dump_req(const struct nlmsghdr *nlh, struct nlattr **tb, 3471 struct nh_dump_filter *filter, 3472 struct netlink_ext_ack *extack) 3473{ 3474 struct nhmsg *nhm; 3475 u32 idx; 3476 3477 if (tb[NHA_OIF]) { 3478 idx = nla_get_u32(tb[NHA_OIF]); 3479 if (idx > INT_MAX) { 3480 NL_SET_ERR_MSG(extack, "Invalid device index"); 3481 return -EINVAL; 3482 } 3483 filter->dev_idx = idx; 3484 } 3485 if (tb[NHA_MASTER]) { 3486 idx = nla_get_u32(tb[NHA_MASTER]); 3487 if (idx > INT_MAX) { 3488 NL_SET_ERR_MSG(extack, "Invalid master device index"); 3489 return -EINVAL; 3490 } 3491 filter->master_idx = idx; 3492 } 3493 filter->group_filter = nla_get_flag(tb[NHA_GROUPS]); 3494 filter->fdb_filter = nla_get_flag(tb[NHA_FDB]); 3495 3496 nhm = nlmsg_data(nlh); 3497 if (nhm->nh_protocol || nhm->resvd || nhm->nh_scope || nhm->nh_flags) { 3498 NL_SET_ERR_MSG(extack, "Invalid values in header for nexthop dump request"); 3499 return -EINVAL; 3500 } 3501 3502 return 0; 3503} 3504 3505static int nh_valid_dump_req(const struct nlmsghdr *nlh, 3506 struct nh_dump_filter *filter, 3507 struct netlink_callback *cb) 3508{ 3509 struct nlattr *tb[ARRAY_SIZE(rtm_nh_policy_dump)]; 3510 int err; 3511 3512 err = nlmsg_parse(nlh, sizeof(struct nhmsg), tb, 3513 ARRAY_SIZE(rtm_nh_policy_dump) - 1, 3514 rtm_nh_policy_dump, cb->extack); 3515 if (err < 0) 3516 return err; 3517 3518 filter->op_flags = nla_get_u32_default(tb[NHA_OP_FLAGS], 0); 3519 3520 return __nh_valid_dump_req(nlh, tb, filter, cb->extack); 3521} 3522 3523struct rtm_dump_nh_ctx { 3524 u32 idx; 3525}; 3526 3527static struct rtm_dump_nh_ctx * 3528rtm_dump_nh_ctx(struct netlink_callback *cb) 3529{ 3530 struct rtm_dump_nh_ctx *ctx = (void *)cb->ctx; 3531 3532 BUILD_BUG_ON(sizeof(*ctx) > sizeof(cb->ctx)); 3533 return ctx; 3534} 3535 3536static int rtm_dump_walk_nexthops(struct sk_buff *skb, 3537 struct netlink_callback *cb, 3538 struct rb_root *root, 3539 struct rtm_dump_nh_ctx *ctx, 3540 int (*nh_cb)(struct sk_buff *skb, 3541 struct netlink_callback *cb, 3542 struct nexthop *nh, void *data), 3543 void *data) 3544{ 3545 struct rb_node *node; 3546 int s_idx; 3547 int err; 3548 3549 s_idx = ctx->idx; 3550 3551 /* If this is not the first invocation, ctx->idx will contain the id of 3552 * the last nexthop we processed. Instead of starting from the very 3553 * first element of the red/black tree again and linearly skipping the 3554 * (potentially large) set of nodes with an id smaller than s_idx, walk 3555 * the tree and find the left-most node whose id is >= s_idx. This 3556 * provides an efficient O(log n) starting point for the dump 3557 * continuation. 3558 */ 3559 if (s_idx != 0) { 3560 struct rb_node *tmp = root->rb_node; 3561 3562 node = NULL; 3563 while (tmp) { 3564 struct nexthop *nh; 3565 3566 nh = rb_entry(tmp, struct nexthop, rb_node); 3567 if (nh->id < s_idx) { 3568 tmp = tmp->rb_right; 3569 } else { 3570 /* Track current candidate and keep looking on 3571 * the left side to find the left-most 3572 * (smallest id) that is still >= s_idx. 3573 */ 3574 node = tmp; 3575 tmp = tmp->rb_left; 3576 } 3577 } 3578 } else { 3579 node = rb_first(root); 3580 } 3581 3582 for (; node; node = rb_next(node)) { 3583 struct nexthop *nh; 3584 3585 nh = rb_entry(node, struct nexthop, rb_node); 3586 3587 ctx->idx = nh->id; 3588 err = nh_cb(skb, cb, nh, data); 3589 if (err) 3590 return err; 3591 } 3592 3593 return 0; 3594} 3595 3596static int rtm_dump_nexthop_cb(struct sk_buff *skb, struct netlink_callback *cb, 3597 struct nexthop *nh, void *data) 3598{ 3599 struct nhmsg *nhm = nlmsg_data(cb->nlh); 3600 struct nh_dump_filter *filter = data; 3601 3602 if (nh_dump_filtered(nh, filter, nhm->nh_family)) 3603 return 0; 3604 3605 return nh_fill_node(skb, nh, RTM_NEWNEXTHOP, 3606 NETLINK_CB(cb->skb).portid, 3607 cb->nlh->nlmsg_seq, NLM_F_MULTI, filter->op_flags); 3608} 3609 3610/* rtnl */ 3611static int rtm_dump_nexthop(struct sk_buff *skb, struct netlink_callback *cb) 3612{ 3613 struct rtm_dump_nh_ctx *ctx = rtm_dump_nh_ctx(cb); 3614 struct net *net = sock_net(skb->sk); 3615 struct rb_root *root = &net->nexthop.rb_root; 3616 struct nh_dump_filter filter = {}; 3617 int err; 3618 3619 err = nh_valid_dump_req(cb->nlh, &filter, cb); 3620 if (err < 0) 3621 return err; 3622 3623 err = rtm_dump_walk_nexthops(skb, cb, root, ctx, 3624 &rtm_dump_nexthop_cb, &filter); 3625 3626 cb->seq = net->nexthop.seq; 3627 nl_dump_check_consistent(cb, nlmsg_hdr(skb)); 3628 return err; 3629} 3630 3631static struct nexthop * 3632nexthop_find_group_resilient(struct net *net, u32 id, 3633 struct netlink_ext_ack *extack) 3634{ 3635 struct nh_group *nhg; 3636 struct nexthop *nh; 3637 3638 nh = nexthop_find_by_id(net, id); 3639 if (!nh) 3640 return ERR_PTR(-ENOENT); 3641 3642 if (!nh->is_group) { 3643 NL_SET_ERR_MSG(extack, "Not a nexthop group"); 3644 return ERR_PTR(-EINVAL); 3645 } 3646 3647 nhg = rtnl_dereference(nh->nh_grp); 3648 if (!nhg->resilient) { 3649 NL_SET_ERR_MSG(extack, "Nexthop group not of type resilient"); 3650 return ERR_PTR(-EINVAL); 3651 } 3652 3653 return nh; 3654} 3655 3656static int nh_valid_dump_nhid(struct nlattr *attr, u32 *nh_id_p, 3657 struct netlink_ext_ack *extack) 3658{ 3659 u32 idx; 3660 3661 if (attr) { 3662 idx = nla_get_u32(attr); 3663 if (!idx) { 3664 NL_SET_ERR_MSG(extack, "Invalid nexthop id"); 3665 return -EINVAL; 3666 } 3667 *nh_id_p = idx; 3668 } else { 3669 *nh_id_p = 0; 3670 } 3671 3672 return 0; 3673} 3674 3675static int nh_valid_dump_bucket_req(const struct nlmsghdr *nlh, 3676 struct nh_dump_filter *filter, 3677 struct netlink_callback *cb) 3678{ 3679 struct nlattr *res_tb[ARRAY_SIZE(rtm_nh_res_bucket_policy_dump)]; 3680 struct nlattr *tb[ARRAY_SIZE(rtm_nh_policy_dump_bucket)]; 3681 int err; 3682 3683 err = nlmsg_parse(nlh, sizeof(struct nhmsg), tb, 3684 ARRAY_SIZE(rtm_nh_policy_dump_bucket) - 1, 3685 rtm_nh_policy_dump_bucket, NULL); 3686 if (err < 0) 3687 return err; 3688 3689 err = nh_valid_dump_nhid(tb[NHA_ID], &filter->nh_id, cb->extack); 3690 if (err) 3691 return err; 3692 3693 if (tb[NHA_RES_BUCKET]) { 3694 size_t max = ARRAY_SIZE(rtm_nh_res_bucket_policy_dump) - 1; 3695 3696 err = nla_parse_nested(res_tb, max, 3697 tb[NHA_RES_BUCKET], 3698 rtm_nh_res_bucket_policy_dump, 3699 cb->extack); 3700 if (err < 0) 3701 return err; 3702 3703 err = nh_valid_dump_nhid(res_tb[NHA_RES_BUCKET_NH_ID], 3704 &filter->res_bucket_nh_id, 3705 cb->extack); 3706 if (err) 3707 return err; 3708 } 3709 3710 return __nh_valid_dump_req(nlh, tb, filter, cb->extack); 3711} 3712 3713struct rtm_dump_res_bucket_ctx { 3714 struct rtm_dump_nh_ctx nh; 3715 u16 bucket_index; 3716}; 3717 3718static struct rtm_dump_res_bucket_ctx * 3719rtm_dump_res_bucket_ctx(struct netlink_callback *cb) 3720{ 3721 struct rtm_dump_res_bucket_ctx *ctx = (void *)cb->ctx; 3722 3723 BUILD_BUG_ON(sizeof(*ctx) > sizeof(cb->ctx)); 3724 return ctx; 3725} 3726 3727struct rtm_dump_nexthop_bucket_data { 3728 struct rtm_dump_res_bucket_ctx *ctx; 3729 struct nh_dump_filter filter; 3730}; 3731 3732static int rtm_dump_nexthop_bucket_nh(struct sk_buff *skb, 3733 struct netlink_callback *cb, 3734 struct nexthop *nh, 3735 struct rtm_dump_nexthop_bucket_data *dd) 3736{ 3737 u32 portid = NETLINK_CB(cb->skb).portid; 3738 struct nhmsg *nhm = nlmsg_data(cb->nlh); 3739 struct nh_res_table *res_table; 3740 struct nh_group *nhg; 3741 u16 bucket_index; 3742 int err; 3743 3744 nhg = rtnl_dereference(nh->nh_grp); 3745 res_table = rtnl_dereference(nhg->res_table); 3746 for (bucket_index = dd->ctx->bucket_index; 3747 bucket_index < res_table->num_nh_buckets; 3748 bucket_index++) { 3749 struct nh_res_bucket *bucket; 3750 struct nh_grp_entry *nhge; 3751 3752 bucket = &res_table->nh_buckets[bucket_index]; 3753 nhge = rtnl_dereference(bucket->nh_entry); 3754 if (nh_dump_filtered(nhge->nh, &dd->filter, nhm->nh_family)) 3755 continue; 3756 3757 if (dd->filter.res_bucket_nh_id && 3758 dd->filter.res_bucket_nh_id != nhge->nh->id) 3759 continue; 3760 3761 dd->ctx->bucket_index = bucket_index; 3762 err = nh_fill_res_bucket(skb, nh, bucket, bucket_index, 3763 RTM_NEWNEXTHOPBUCKET, portid, 3764 cb->nlh->nlmsg_seq, NLM_F_MULTI, 3765 cb->extack); 3766 if (err) 3767 return err; 3768 } 3769 3770 dd->ctx->bucket_index = 0; 3771 3772 return 0; 3773} 3774 3775static int rtm_dump_nexthop_bucket_cb(struct sk_buff *skb, 3776 struct netlink_callback *cb, 3777 struct nexthop *nh, void *data) 3778{ 3779 struct rtm_dump_nexthop_bucket_data *dd = data; 3780 struct nh_group *nhg; 3781 3782 if (!nh->is_group) 3783 return 0; 3784 3785 nhg = rtnl_dereference(nh->nh_grp); 3786 if (!nhg->resilient) 3787 return 0; 3788 3789 return rtm_dump_nexthop_bucket_nh(skb, cb, nh, dd); 3790} 3791 3792/* rtnl */ 3793static int rtm_dump_nexthop_bucket(struct sk_buff *skb, 3794 struct netlink_callback *cb) 3795{ 3796 struct rtm_dump_res_bucket_ctx *ctx = rtm_dump_res_bucket_ctx(cb); 3797 struct rtm_dump_nexthop_bucket_data dd = { .ctx = ctx }; 3798 struct net *net = sock_net(skb->sk); 3799 struct nexthop *nh; 3800 int err; 3801 3802 err = nh_valid_dump_bucket_req(cb->nlh, &dd.filter, cb); 3803 if (err) 3804 return err; 3805 3806 if (dd.filter.nh_id) { 3807 nh = nexthop_find_group_resilient(net, dd.filter.nh_id, 3808 cb->extack); 3809 if (IS_ERR(nh)) 3810 return PTR_ERR(nh); 3811 err = rtm_dump_nexthop_bucket_nh(skb, cb, nh, &dd); 3812 } else { 3813 struct rb_root *root = &net->nexthop.rb_root; 3814 3815 err = rtm_dump_walk_nexthops(skb, cb, root, &ctx->nh, 3816 &rtm_dump_nexthop_bucket_cb, &dd); 3817 } 3818 3819 cb->seq = net->nexthop.seq; 3820 nl_dump_check_consistent(cb, nlmsg_hdr(skb)); 3821 return err; 3822} 3823 3824static int nh_valid_get_bucket_req_res_bucket(struct nlattr *res, 3825 u16 *bucket_index, 3826 struct netlink_ext_ack *extack) 3827{ 3828 struct nlattr *tb[ARRAY_SIZE(rtm_nh_res_bucket_policy_get)]; 3829 int err; 3830 3831 err = nla_parse_nested(tb, ARRAY_SIZE(rtm_nh_res_bucket_policy_get) - 1, 3832 res, rtm_nh_res_bucket_policy_get, extack); 3833 if (err < 0) 3834 return err; 3835 3836 if (!tb[NHA_RES_BUCKET_INDEX]) { 3837 NL_SET_ERR_MSG(extack, "Bucket index is missing"); 3838 return -EINVAL; 3839 } 3840 3841 *bucket_index = nla_get_u16(tb[NHA_RES_BUCKET_INDEX]); 3842 return 0; 3843} 3844 3845static int nh_valid_get_bucket_req(const struct nlmsghdr *nlh, 3846 u32 *id, u16 *bucket_index, 3847 struct netlink_ext_ack *extack) 3848{ 3849 struct nlattr *tb[ARRAY_SIZE(rtm_nh_policy_get_bucket)]; 3850 int err; 3851 3852 err = nlmsg_parse(nlh, sizeof(struct nhmsg), tb, 3853 ARRAY_SIZE(rtm_nh_policy_get_bucket) - 1, 3854 rtm_nh_policy_get_bucket, extack); 3855 if (err < 0) 3856 return err; 3857 3858 err = nh_valid_get_del_req(nlh, tb, id, NULL, extack); 3859 if (err) 3860 return err; 3861 3862 if (!tb[NHA_RES_BUCKET]) { 3863 NL_SET_ERR_MSG(extack, "Bucket information is missing"); 3864 return -EINVAL; 3865 } 3866 3867 err = nh_valid_get_bucket_req_res_bucket(tb[NHA_RES_BUCKET], 3868 bucket_index, extack); 3869 if (err) 3870 return err; 3871 3872 return 0; 3873} 3874 3875/* rtnl */ 3876static int rtm_get_nexthop_bucket(struct sk_buff *in_skb, struct nlmsghdr *nlh, 3877 struct netlink_ext_ack *extack) 3878{ 3879 struct net *net = sock_net(in_skb->sk); 3880 struct nh_res_table *res_table; 3881 struct sk_buff *skb = NULL; 3882 struct nh_group *nhg; 3883 struct nexthop *nh; 3884 u16 bucket_index; 3885 int err; 3886 u32 id; 3887 3888 err = nh_valid_get_bucket_req(nlh, &id, &bucket_index, extack); 3889 if (err) 3890 return err; 3891 3892 nh = nexthop_find_group_resilient(net, id, extack); 3893 if (IS_ERR(nh)) 3894 return PTR_ERR(nh); 3895 3896 nhg = rtnl_dereference(nh->nh_grp); 3897 res_table = rtnl_dereference(nhg->res_table); 3898 if (bucket_index >= res_table->num_nh_buckets) { 3899 NL_SET_ERR_MSG(extack, "Bucket index out of bounds"); 3900 return -ENOENT; 3901 } 3902 3903 skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); 3904 if (!skb) 3905 return -ENOBUFS; 3906 3907 err = nh_fill_res_bucket(skb, nh, &res_table->nh_buckets[bucket_index], 3908 bucket_index, RTM_NEWNEXTHOPBUCKET, 3909 NETLINK_CB(in_skb).portid, nlh->nlmsg_seq, 3910 0, extack); 3911 if (err < 0) { 3912 WARN_ON(err == -EMSGSIZE); 3913 goto errout_free; 3914 } 3915 3916 return rtnl_unicast(skb, net, NETLINK_CB(in_skb).portid); 3917 3918errout_free: 3919 kfree_skb(skb); 3920 return err; 3921} 3922 3923static void nexthop_sync_mtu(struct net_device *dev, u32 orig_mtu) 3924{ 3925 unsigned int hash = nh_dev_hashfn(dev->ifindex); 3926 struct net *net = dev_net(dev); 3927 struct hlist_head *head = &net->nexthop.devhash[hash]; 3928 struct hlist_node *n; 3929 struct nh_info *nhi; 3930 3931 hlist_for_each_entry_safe(nhi, n, head, dev_hash) { 3932 if (nhi->fib_nhc.nhc_dev == dev) { 3933 if (nhi->family == AF_INET) 3934 fib_nhc_update_mtu(&nhi->fib_nhc, dev->mtu, 3935 orig_mtu); 3936 } 3937 } 3938} 3939 3940/* rtnl */ 3941static int nh_netdev_event(struct notifier_block *this, 3942 unsigned long event, void *ptr) 3943{ 3944 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 3945 struct netdev_notifier_info_ext *info_ext; 3946 3947 switch (event) { 3948 case NETDEV_DOWN: 3949 case NETDEV_UNREGISTER: 3950 nexthop_flush_dev(dev, event); 3951 break; 3952 case NETDEV_CHANGE: 3953 if (!(netif_get_flags(dev) & (IFF_RUNNING | IFF_LOWER_UP))) 3954 nexthop_flush_dev(dev, event); 3955 break; 3956 case NETDEV_CHANGEMTU: 3957 info_ext = ptr; 3958 nexthop_sync_mtu(dev, info_ext->ext.mtu); 3959 rt_cache_flush(dev_net(dev)); 3960 break; 3961 } 3962 return NOTIFY_DONE; 3963} 3964 3965static struct notifier_block nh_netdev_notifier = { 3966 .notifier_call = nh_netdev_event, 3967}; 3968 3969static int nexthops_dump(struct net *net, struct notifier_block *nb, 3970 enum nexthop_event_type event_type, 3971 struct netlink_ext_ack *extack) 3972{ 3973 struct rb_root *root = &net->nexthop.rb_root; 3974 struct rb_node *node; 3975 int err = 0; 3976 3977 for (node = rb_first(root); node; node = rb_next(node)) { 3978 struct nexthop *nh; 3979 3980 nh = rb_entry(node, struct nexthop, rb_node); 3981 err = call_nexthop_notifier(nb, net, event_type, nh, extack); 3982 if (err) 3983 break; 3984 } 3985 3986 return err; 3987} 3988 3989int register_nexthop_notifier(struct net *net, struct notifier_block *nb, 3990 struct netlink_ext_ack *extack) 3991{ 3992 int err; 3993 3994 rtnl_lock(); 3995 err = nexthops_dump(net, nb, NEXTHOP_EVENT_REPLACE, extack); 3996 if (err) 3997 goto unlock; 3998 err = blocking_notifier_chain_register(&net->nexthop.notifier_chain, 3999 nb); 4000unlock: 4001 rtnl_unlock(); 4002 return err; 4003} 4004EXPORT_SYMBOL(register_nexthop_notifier); 4005 4006int __unregister_nexthop_notifier(struct net *net, struct notifier_block *nb) 4007{ 4008 int err; 4009 4010 err = blocking_notifier_chain_unregister(&net->nexthop.notifier_chain, 4011 nb); 4012 if (!err) 4013 nexthops_dump(net, nb, NEXTHOP_EVENT_DEL, NULL); 4014 return err; 4015} 4016EXPORT_SYMBOL(__unregister_nexthop_notifier); 4017 4018int unregister_nexthop_notifier(struct net *net, struct notifier_block *nb) 4019{ 4020 int err; 4021 4022 rtnl_lock(); 4023 err = __unregister_nexthop_notifier(net, nb); 4024 rtnl_unlock(); 4025 return err; 4026} 4027EXPORT_SYMBOL(unregister_nexthop_notifier); 4028 4029void nexthop_set_hw_flags(struct net *net, u32 id, bool offload, bool trap) 4030{ 4031 struct nexthop *nexthop; 4032 4033 rcu_read_lock(); 4034 4035 nexthop = nexthop_find_by_id(net, id); 4036 if (!nexthop) 4037 goto out; 4038 4039 nexthop->nh_flags &= ~(RTNH_F_OFFLOAD | RTNH_F_TRAP); 4040 if (offload) 4041 nexthop->nh_flags |= RTNH_F_OFFLOAD; 4042 if (trap) 4043 nexthop->nh_flags |= RTNH_F_TRAP; 4044 4045out: 4046 rcu_read_unlock(); 4047} 4048EXPORT_SYMBOL(nexthop_set_hw_flags); 4049 4050void nexthop_bucket_set_hw_flags(struct net *net, u32 id, u16 bucket_index, 4051 bool offload, bool trap) 4052{ 4053 struct nh_res_table *res_table; 4054 struct nh_res_bucket *bucket; 4055 struct nexthop *nexthop; 4056 struct nh_group *nhg; 4057 4058 rcu_read_lock(); 4059 4060 nexthop = nexthop_find_by_id(net, id); 4061 if (!nexthop || !nexthop->is_group) 4062 goto out; 4063 4064 nhg = rcu_dereference(nexthop->nh_grp); 4065 if (!nhg->resilient) 4066 goto out; 4067 4068 if (bucket_index >= nhg->res_table->num_nh_buckets) 4069 goto out; 4070 4071 res_table = rcu_dereference(nhg->res_table); 4072 bucket = &res_table->nh_buckets[bucket_index]; 4073 bucket->nh_flags &= ~(RTNH_F_OFFLOAD | RTNH_F_TRAP); 4074 if (offload) 4075 bucket->nh_flags |= RTNH_F_OFFLOAD; 4076 if (trap) 4077 bucket->nh_flags |= RTNH_F_TRAP; 4078 4079out: 4080 rcu_read_unlock(); 4081} 4082EXPORT_SYMBOL(nexthop_bucket_set_hw_flags); 4083 4084void nexthop_res_grp_activity_update(struct net *net, u32 id, u16 num_buckets, 4085 unsigned long *activity) 4086{ 4087 struct nh_res_table *res_table; 4088 struct nexthop *nexthop; 4089 struct nh_group *nhg; 4090 u16 i; 4091 4092 rcu_read_lock(); 4093 4094 nexthop = nexthop_find_by_id(net, id); 4095 if (!nexthop || !nexthop->is_group) 4096 goto out; 4097 4098 nhg = rcu_dereference(nexthop->nh_grp); 4099 if (!nhg->resilient) 4100 goto out; 4101 4102 /* Instead of silently ignoring some buckets, demand that the sizes 4103 * be the same. 4104 */ 4105 res_table = rcu_dereference(nhg->res_table); 4106 if (num_buckets != res_table->num_nh_buckets) 4107 goto out; 4108 4109 for (i = 0; i < num_buckets; i++) { 4110 if (test_bit(i, activity)) 4111 nh_res_bucket_set_busy(&res_table->nh_buckets[i]); 4112 } 4113 4114out: 4115 rcu_read_unlock(); 4116} 4117EXPORT_SYMBOL(nexthop_res_grp_activity_update); 4118 4119static void __net_exit nexthop_net_exit_rtnl(struct net *net, 4120 struct list_head *dev_to_kill) 4121{ 4122 ASSERT_RTNL_NET(net); 4123 flush_all_nexthops(net); 4124} 4125 4126static void __net_exit nexthop_net_exit(struct net *net) 4127{ 4128 kfree(net->nexthop.devhash); 4129 net->nexthop.devhash = NULL; 4130} 4131 4132static int __net_init nexthop_net_init(struct net *net) 4133{ 4134 size_t sz = sizeof(struct hlist_head) * NH_DEV_HASHSIZE; 4135 4136 net->nexthop.rb_root = RB_ROOT; 4137 net->nexthop.devhash = kzalloc(sz, GFP_KERNEL); 4138 if (!net->nexthop.devhash) 4139 return -ENOMEM; 4140 BLOCKING_INIT_NOTIFIER_HEAD(&net->nexthop.notifier_chain); 4141 4142 return 0; 4143} 4144 4145static struct pernet_operations nexthop_net_ops = { 4146 .init = nexthop_net_init, 4147 .exit = nexthop_net_exit, 4148 .exit_rtnl = nexthop_net_exit_rtnl, 4149}; 4150 4151static const struct rtnl_msg_handler nexthop_rtnl_msg_handlers[] __initconst = { 4152 {.msgtype = RTM_NEWNEXTHOP, .doit = rtm_new_nexthop, 4153 .flags = RTNL_FLAG_DOIT_PERNET}, 4154 {.msgtype = RTM_DELNEXTHOP, .doit = rtm_del_nexthop, 4155 .flags = RTNL_FLAG_DOIT_PERNET}, 4156 {.msgtype = RTM_GETNEXTHOP, .doit = rtm_get_nexthop, 4157 .dumpit = rtm_dump_nexthop}, 4158 {.msgtype = RTM_GETNEXTHOPBUCKET, .doit = rtm_get_nexthop_bucket, 4159 .dumpit = rtm_dump_nexthop_bucket}, 4160 {.protocol = PF_INET, .msgtype = RTM_NEWNEXTHOP, 4161 .doit = rtm_new_nexthop, .flags = RTNL_FLAG_DOIT_PERNET}, 4162 {.protocol = PF_INET, .msgtype = RTM_GETNEXTHOP, 4163 .dumpit = rtm_dump_nexthop}, 4164 {.protocol = PF_INET6, .msgtype = RTM_NEWNEXTHOP, 4165 .doit = rtm_new_nexthop, .flags = RTNL_FLAG_DOIT_PERNET}, 4166 {.protocol = PF_INET6, .msgtype = RTM_GETNEXTHOP, 4167 .dumpit = rtm_dump_nexthop}, 4168}; 4169 4170static int __init nexthop_init(void) 4171{ 4172 register_pernet_subsys(&nexthop_net_ops); 4173 4174 register_netdevice_notifier(&nh_netdev_notifier); 4175 4176 rtnl_register_many(nexthop_rtnl_msg_handlers); 4177 4178 return 0; 4179} 4180subsys_initcall(nexthop_init);
~/
Read-only file preview.