net/mctp/route.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Management Component Transport Protocol (MCTP) - routing
   4  * implementation.
   5  *
   6  * This is currently based on a simple routing table, with no dst cache. The
   7  * number of routes should stay fairly small, so the lookup cost is small.
   8  *
   9  * Copyright (c) 2021 Code Construct
  10  * Copyright (c) 2021 Google
  11  */
  12
  13 #include <linux/idr.h>
  14 #include <linux/kconfig.h>
  15 #include <linux/mctp.h>
  16 #include <linux/netdevice.h>
  17 #include <linux/rtnetlink.h>
  18 #include <linux/skbuff.h>
  19
  20 #include <uapi/linux/if_arp.h>
  21
  22 #include <net/mctp.h>
  23 #include <net/mctpdevice.h>
  24 #include <net/netlink.h>
  25 #include <net/sock.h>
  26
  27 #include <trace/events/mctp.h>
  28
  29 static const unsigned int mctp_message_maxlen = 64 * 1024;
  30 static const unsigned long mctp_key_lifetime = 6 * CONFIG_HZ;
  31
  32 static void mctp_flow_prepare_output(struct sk_buff *skb, struct mctp_dev *dev);
  33
  34 /* route output callbacks */
  35 static int mctp_route_discard(struct mctp_route *route, struct sk_buff *skb)
  36 {
  37         kfree_skb(skb);
  38         return 0;
  39 }
  40
  41 static struct mctp_sock *mctp_lookup_bind(struct net *net, struct sk_buff *skb)
  42 {
  43         struct mctp_skb_cb *cb = mctp_cb(skb);
  44         struct mctp_hdr *mh;
  45         struct sock *sk;
  46         u8 type;
  47
  48         WARN_ON(!rcu_read_lock_held());
  49
  50         /* TODO: look up in skb->cb? */
  51         mh = mctp_hdr(skb);
  52
  53         if (!skb_headlen(skb))
  54                 return NULL;
  55
  56         type = (*(u8 *)skb->data) & 0x7f;
  57
  58         sk_for_each_rcu(sk, &net->mctp.binds) {
  59                 struct mctp_sock *msk = container_of(sk, struct mctp_sock, sk);
  60
  61                 if (msk->bind_net != MCTP_NET_ANY && msk->bind_net != cb->net)
  62                         continue;
  63
  64                 if (msk->bind_type != type)
  65                         continue;
  66
  67                 if (!mctp_address_matches(msk->bind_addr, mh->dest))
  68                         continue;
  69
  70                 return msk;
  71         }
  72
  73         return NULL;
  74 }
  75
  76 static bool mctp_key_match(struct mctp_sk_key *key, mctp_eid_t local,
  77                            mctp_eid_t peer, u8 tag)
  78 {
  79         if (!mctp_address_matches(key->local_addr, local))
  80                 return false;
  81
  82         if (key->peer_addr != peer)
  83                 return false;
  84
  85         if (key->tag != tag)
  86                 return false;
  87
  88         return true;
  89 }
  90
  91 /* returns a key (with key->lock held, and refcounted), or NULL if no such
  92  * key exists.
  93  */
  94 static struct mctp_sk_key *mctp_lookup_key(struct net *net, struct sk_buff *skb,
  95                                            mctp_eid_t peer,
  96                                            unsigned long *irqflags)
  97         __acquires(&key->lock)
  98 {
  99         struct mctp_sk_key *key, *ret;
 100         unsigned long flags;
 101         struct mctp_hdr *mh;
 102         u8 tag;
 103
 104         mh = mctp_hdr(skb);
 105         tag = mh->flags_seq_tag & (MCTP_HDR_TAG_MASK | MCTP_HDR_FLAG_TO);
 106
 107         ret = NULL;
 108         spin_lock_irqsave(&net->mctp.keys_lock, flags);
 109
 110         hlist_for_each_entry(key, &net->mctp.keys, hlist) {
 111                 if (!mctp_key_match(key, mh->dest, peer, tag))
 112                         continue;
 113
 114                 spin_lock(&key->lock);
 115                 if (key->valid) {
 116                         refcount_inc(&key->refs);
 117                         ret = key;
 118                         break;
 119                 }
 120                 spin_unlock(&key->lock);
 121         }
 122
 123         if (ret) {
 124                 spin_unlock(&net->mctp.keys_lock);
 125                 *irqflags = flags;
 126         } else {
 127                 spin_unlock_irqrestore(&net->mctp.keys_lock, flags);
 128         }
 129
 130         return ret;
 131 }
 132
 133 static struct mctp_sk_key *mctp_key_alloc(struct mctp_sock *msk,
 134                                           mctp_eid_t local, mctp_eid_t peer,
 135                                           u8 tag, gfp_t gfp)
 136 {
 137         struct mctp_sk_key *key;
 138
 139         key = kzalloc(sizeof(*key), gfp);
 140         if (!key)
 141                 return NULL;
 142
 143         key->peer_addr = peer;
 144         key->local_addr = local;
 145         key->tag = tag;
 146         key->sk = &msk->sk;
 147         key->valid = true;
 148         spin_lock_init(&key->lock);
 149         refcount_set(&key->refs, 1);
 150         sock_hold(key->sk);
 151
 152         return key;
 153 }
 154
 155 void mctp_key_unref(struct mctp_sk_key *key)
 156 {
 157         unsigned long flags;
 158
 159         if (!refcount_dec_and_test(&key->refs))
 160                 return;
 161
 162         /* even though no refs exist here, the lock allows us to stay
 163          * consistent with the locking requirement of mctp_dev_release_key
 164          */
 165         spin_lock_irqsave(&key->lock, flags);
 166         mctp_dev_release_key(key->dev, key);
 167         spin_unlock_irqrestore(&key->lock, flags);
 168
 169         sock_put(key->sk);
 170         kfree(key);
 171 }
 172
 173 static int mctp_key_add(struct mctp_sk_key *key, struct mctp_sock *msk)
 174 {
 175         struct net *net = sock_net(&msk->sk);
 176         struct mctp_sk_key *tmp;
 177         unsigned long flags;
 178         int rc = 0;
 179
 180         spin_lock_irqsave(&net->mctp.keys_lock, flags);
 181
 182         if (sock_flag(&msk->sk, SOCK_DEAD)) {
 183                 rc = -EINVAL;
 184                 goto out_unlock;
 185         }
 186
 187         hlist_for_each_entry(tmp, &net->mctp.keys, hlist) {
 188                 if (mctp_key_match(tmp, key->local_addr, key->peer_addr,
 189                                    key->tag)) {
 190                         spin_lock(&tmp->lock);
 191                         if (tmp->valid)
 192                                 rc = -EEXIST;
 193                         spin_unlock(&tmp->lock);
 194                         if (rc)
 195                                 break;
 196                 }
 197         }
 198
 199         if (!rc) {
 200                 refcount_inc(&key->refs);
 201                 key->expiry = jiffies + mctp_key_lifetime;
 202                 timer_reduce(&msk->key_expiry, key->expiry);
 203
 204                 hlist_add_head(&key->hlist, &net->mctp.keys);
 205                 hlist_add_head(&key->sklist, &msk->keys);
 206         }
 207
 208 out_unlock:
 209         spin_unlock_irqrestore(&net->mctp.keys_lock, flags);
 210
 211         return rc;
 212 }
 213
 214 /* Helper for mctp_route_input().
 215  * We're done with the key; unlock and unref the key.
 216  * For the usual case of automatic expiry we remove the key from lists.
 217  * In the case that manual allocation is set on a key we release the lock
 218  * and local ref, reset reassembly, but don't remove from lists.
 219  */
 220 static void __mctp_key_done_in(struct mctp_sk_key *key, struct net *net,
 221                                unsigned long flags, unsigned long reason)
 222 __releases(&key->lock)
 223 {
 224         struct sk_buff *skb;
 225
 226         trace_mctp_key_release(key, reason);
 227         skb = key->reasm_head;
 228         key->reasm_head = NULL;
 229
 230         if (!key->manual_alloc) {
 231                 key->reasm_dead = true;
 232                 key->valid = false;
 233                 mctp_dev_release_key(key->dev, key);
 234         }
 235         spin_unlock_irqrestore(&key->lock, flags);
 236
 237         if (!key->manual_alloc) {
 238                 spin_lock_irqsave(&net->mctp.keys_lock, flags);
 239                 if (!hlist_unhashed(&key->hlist)) {
 240                         hlist_del_init(&key->hlist);
 241                         hlist_del_init(&key->sklist);
 242                         mctp_key_unref(key);
 243                 }
 244                 spin_unlock_irqrestore(&net->mctp.keys_lock, flags);
 245         }
 246
 247         /* and one for the local reference */
 248         mctp_key_unref(key);
 249
 250         kfree_skb(skb);
 251 }
 252
 253 #ifdef CONFIG_MCTP_FLOWS
 254 static void mctp_skb_set_flow(struct sk_buff *skb, struct mctp_sk_key *key)
 255 {
 256         struct mctp_flow *flow;
 257
 258         flow = skb_ext_add(skb, SKB_EXT_MCTP);
 259         if (!flow)
 260                 return;
 261
 262         refcount_inc(&key->refs);
 263         flow->key = key;
 264 }
 265
 266 static void mctp_flow_prepare_output(struct sk_buff *skb, struct mctp_dev *dev)
 267 {
 268         struct mctp_sk_key *key;
 269         struct mctp_flow *flow;
 270
 271         flow = skb_ext_find(skb, SKB_EXT_MCTP);
 272         if (!flow)
 273                 return;
 274
 275         key = flow->key;
 276
 277         if (WARN_ON(key->dev && key->dev != dev))
 278                 return;
 279
 280         mctp_dev_set_key(dev, key);
 281 }
 282 #else
 283 static void mctp_skb_set_flow(struct sk_buff *skb, struct mctp_sk_key *key) {}
 284 static void mctp_flow_prepare_output(struct sk_buff *skb, struct mctp_dev *dev) {}
 285 #endif
 286
 287 static int mctp_frag_queue(struct mctp_sk_key *key, struct sk_buff *skb)
 288 {
 289         struct mctp_hdr *hdr = mctp_hdr(skb);
 290         u8 exp_seq, this_seq;
 291
 292         this_seq = (hdr->flags_seq_tag >> MCTP_HDR_SEQ_SHIFT)
 293                 & MCTP_HDR_SEQ_MASK;
 294
 295         if (!key->reasm_head) {
 296                 key->reasm_head = skb;
 297                 key->reasm_tailp = &(skb_shinfo(skb)->frag_list);
 298                 key->last_seq = this_seq;
 299                 return 0;
 300         }
 301
 302         exp_seq = (key->last_seq + 1) & MCTP_HDR_SEQ_MASK;
 303
 304         if (this_seq != exp_seq)
 305                 return -EINVAL;
 306
 307         if (key->reasm_head->len + skb->len > mctp_message_maxlen)
 308                 return -EINVAL;
 309
 310         skb->next = NULL;
 311         skb->sk = NULL;
 312         *key->reasm_tailp = skb;
 313         key->reasm_tailp = &skb->next;
 314
 315         key->last_seq = this_seq;
 316
 317         key->reasm_head->data_len += skb->len;
 318         key->reasm_head->len += skb->len;
 319         key->reasm_head->truesize += skb->truesize;
 320
 321         return 0;
 322 }
 323
 324 static int mctp_route_input(struct mctp_route *route, struct sk_buff *skb)
 325 {
 326         struct mctp_sk_key *key, *any_key = NULL;
 327         struct net *net = dev_net(skb->dev);
 328         struct mctp_sock *msk;
 329         struct mctp_hdr *mh;
 330         unsigned long f;
 331         u8 tag, flags;
 332         int rc;
 333
 334         msk = NULL;
 335         rc = -EINVAL;
 336
 337         /* we may be receiving a locally-routed packet; drop source sk
 338          * accounting
 339          */
 340         skb_orphan(skb);
 341
 342         /* ensure we have enough data for a header and a type */
 343         if (skb->len < sizeof(struct mctp_hdr) + 1)
 344                 goto out;
 345
 346         /* grab header, advance data ptr */
 347         mh = mctp_hdr(skb);
 348         skb_pull(skb, sizeof(struct mctp_hdr));
 349
 350         if (mh->ver != 1)
 351                 goto out;
 352
 353         flags = mh->flags_seq_tag & (MCTP_HDR_FLAG_SOM | MCTP_HDR_FLAG_EOM);
 354         tag = mh->flags_seq_tag & (MCTP_HDR_TAG_MASK | MCTP_HDR_FLAG_TO);
 355
 356         rcu_read_lock();
 357
 358         /* lookup socket / reasm context, exactly matching (src,dest,tag).
 359          * we hold a ref on the key, and key->lock held.
 360          */
 361         key = mctp_lookup_key(net, skb, mh->src, &f);
 362
 363         if (flags & MCTP_HDR_FLAG_SOM) {
 364                 if (key) {
 365                         msk = container_of(key->sk, struct mctp_sock, sk);
 366                 } else {
 367                         /* first response to a broadcast? do a more general
 368                          * key lookup to find the socket, but don't use this
 369                          * key for reassembly - we'll create a more specific
 370                          * one for future packets if required (ie, !EOM).
 371                          */
 372                         any_key = mctp_lookup_key(net, skb, MCTP_ADDR_ANY, &f);
 373                         if (any_key) {
 374                                 msk = container_of(any_key->sk,
 375                                                    struct mctp_sock, sk);
 376                                 spin_unlock_irqrestore(&any_key->lock, f);
 377                         }
 378                 }
 379
 380                 if (!key && !msk && (tag & MCTP_HDR_FLAG_TO))
 381                         msk = mctp_lookup_bind(net, skb);
 382
 383                 if (!msk) {
 384                         rc = -ENOENT;
 385                         goto out_unlock;
 386                 }
 387
 388                 /* single-packet message? deliver to socket, clean up any
 389                  * pending key.
 390                  */
 391                 if (flags & MCTP_HDR_FLAG_EOM) {
 392                         sock_queue_rcv_skb(&msk->sk, skb);
 393                         if (key) {
 394                                 /* we've hit a pending reassembly; not much we
 395                                  * can do but drop it
 396                                  */
 397                                 __mctp_key_done_in(key, net, f,
 398                                                    MCTP_TRACE_KEY_REPLIED);
 399                                 key = NULL;
 400                         }
 401                         rc = 0;
 402                         goto out_unlock;
 403                 }
 404
 405                 /* broadcast response or a bind() - create a key for further
 406                  * packets for this message
 407                  */
 408                 if (!key) {
 409                         key = mctp_key_alloc(msk, mh->dest, mh->src,
 410                                              tag, GFP_ATOMIC);
 411                         if (!key) {
 412                                 rc = -ENOMEM;
 413                                 goto out_unlock;
 414                         }
 415
 416                         /* we can queue without the key lock here, as the
 417                          * key isn't observable yet
 418                          */
 419                         mctp_frag_queue(key, skb);
 420
 421                         /* if the key_add fails, we've raced with another
 422                          * SOM packet with the same src, dest and tag. There's
 423                          * no way to distinguish future packets, so all we
 424                          * can do is drop; we'll free the skb on exit from
 425                          * this function.
 426                          */
 427                         rc = mctp_key_add(key, msk);
 428                         if (!rc)
 429                                 trace_mctp_key_acquire(key);
 430
 431                         /* we don't need to release key->lock on exit, so
 432                          * clean up here and suppress the unlock via
 433                          * setting to NULL
 434                          */
 435                         mctp_key_unref(key);
 436                         key = NULL;
 437
 438                 } else {
 439                         if (key->reasm_head || key->reasm_dead) {
 440                                 /* duplicate start? drop everything */
 441                                 __mctp_key_done_in(key, net, f,
 442                                                    MCTP_TRACE_KEY_INVALIDATED);
 443                                 rc = -EEXIST;
 444                                 key = NULL;
 445                         } else {
 446                                 rc = mctp_frag_queue(key, skb);
 447                         }
 448                 }
 449
 450         } else if (key) {
 451                 /* this packet continues a previous message; reassemble
 452                  * using the message-specific key
 453                  */
 454
 455                 /* we need to be continuing an existing reassembly... */
 456                 if (!key->reasm_head)
 457                         rc = -EINVAL;
 458                 else
 459                         rc = mctp_frag_queue(key, skb);
 460
 461                 /* end of message? deliver to socket, and we're done with
 462                  * the reassembly/response key
 463                  */
 464                 if (!rc && flags & MCTP_HDR_FLAG_EOM) {
 465                         sock_queue_rcv_skb(key->sk, key->reasm_head);
 466                         key->reasm_head = NULL;
 467                         __mctp_key_done_in(key, net, f, MCTP_TRACE_KEY_REPLIED);
 468                         key = NULL;
 469                 }
 470
 471         } else {
 472                 /* not a start, no matching key */
 473                 rc = -ENOENT;
 474         }
 475
 476 out_unlock:
 477         rcu_read_unlock();
 478         if (key) {
 479                 spin_unlock_irqrestore(&key->lock, f);
 480                 mctp_key_unref(key);
 481         }
 482         if (any_key)
 483                 mctp_key_unref(any_key);
 484 out:
 485         if (rc)
 486                 kfree_skb(skb);
 487         return rc;
 488 }
 489
 490 static unsigned int mctp_route_mtu(struct mctp_route *rt)
 491 {
 492         return rt->mtu ?: READ_ONCE(rt->dev->dev->mtu);
 493 }
 494
 495 static int mctp_route_output(struct mctp_route *route, struct sk_buff *skb)
 496 {
 497         struct mctp_skb_cb *cb = mctp_cb(skb);
 498         struct mctp_hdr *hdr = mctp_hdr(skb);
 499         char daddr_buf[MAX_ADDR_LEN];
 500         char *daddr = NULL;
 501         unsigned int mtu;
 502         int rc;
 503
 504         skb->protocol = htons(ETH_P_MCTP);
 505
 506         mtu = READ_ONCE(skb->dev->mtu);
 507         if (skb->len > mtu) {
 508                 kfree_skb(skb);
 509                 return -EMSGSIZE;
 510         }
 511
 512         if (cb->ifindex) {
 513                 /* direct route; use the hwaddr we stashed in sendmsg */
 514                 if (cb->halen != skb->dev->addr_len) {
 515                         /* sanity check, sendmsg should have already caught this */
 516                         kfree_skb(skb);
 517                         return -EMSGSIZE;
 518                 }
 519                 daddr = cb->haddr;
 520         } else {
 521                 /* If lookup fails let the device handle daddr==NULL */
 522                 if (mctp_neigh_lookup(route->dev, hdr->dest, daddr_buf) == 0)
 523                         daddr = daddr_buf;
 524         }
 525
 526         rc = dev_hard_header(skb, skb->dev, ntohs(skb->protocol),
 527                              daddr, skb->dev->dev_addr, skb->len);
 528         if (rc < 0) {
 529                 kfree_skb(skb);
 530                 return -EHOSTUNREACH;
 531         }
 532
 533         mctp_flow_prepare_output(skb, route->dev);
 534
 535         rc = dev_queue_xmit(skb);
 536         if (rc)
 537                 rc = net_xmit_errno(rc);
 538
 539         return rc;
 540 }
 541
 542 /* route alloc/release */
 543 static void mctp_route_release(struct mctp_route *rt)
 544 {
 545         if (refcount_dec_and_test(&rt->refs)) {
 546                 mctp_dev_put(rt->dev);
 547                 kfree_rcu(rt, rcu);
 548         }
 549 }
 550
 551 /* returns a route with the refcount at 1 */
 552 static struct mctp_route *mctp_route_alloc(void)
 553 {
 554         struct mctp_route *rt;
 555
 556         rt = kzalloc(sizeof(*rt), GFP_KERNEL);
 557         if (!rt)
 558                 return NULL;
 559
 560         INIT_LIST_HEAD(&rt->list);
 561         refcount_set(&rt->refs, 1);
 562         rt->output = mctp_route_discard;
 563
 564         return rt;
 565 }
 566
 567 unsigned int mctp_default_net(struct net *net)
 568 {
 569         return READ_ONCE(net->mctp.default_net);
 570 }
 571
 572 int mctp_default_net_set(struct net *net, unsigned int index)
 573 {
 574         if (index == 0)
 575                 return -EINVAL;
 576         WRITE_ONCE(net->mctp.default_net, index);
 577         return 0;
 578 }
 579
 580 /* tag management */
 581 static void mctp_reserve_tag(struct net *net, struct mctp_sk_key *key,
 582                              struct mctp_sock *msk)
 583 {
 584         struct netns_mctp *mns = &net->mctp;
 585
 586         lockdep_assert_held(&mns->keys_lock);
 587
 588         key->expiry = jiffies + mctp_key_lifetime;
 589         timer_reduce(&msk->key_expiry, key->expiry);
 590
 591         /* we hold the net->key_lock here, allowing updates to both
 592          * then net and sk
 593          */
 594         hlist_add_head_rcu(&key->hlist, &mns->keys);
 595         hlist_add_head_rcu(&key->sklist, &msk->keys);
 596         refcount_inc(&key->refs);
 597 }
 598
 599 /* Allocate a locally-owned tag value for (saddr, daddr), and reserve
 600  * it for the socket msk
 601  */
 602 struct mctp_sk_key *mctp_alloc_local_tag(struct mctp_sock *msk,
 603                                          mctp_eid_t daddr, mctp_eid_t saddr,
 604                                          bool manual, u8 *tagp)
 605 {
 606         struct net *net = sock_net(&msk->sk);
 607         struct netns_mctp *mns = &net->mctp;
 608         struct mctp_sk_key *key, *tmp;
 609         unsigned long flags;
 610         u8 tagbits;
 611
 612         /* for NULL destination EIDs, we may get a response from any peer */
 613         if (daddr == MCTP_ADDR_NULL)
 614                 daddr = MCTP_ADDR_ANY;
 615
 616         /* be optimistic, alloc now */
 617         key = mctp_key_alloc(msk, saddr, daddr, 0, GFP_KERNEL);
 618         if (!key)
 619                 return ERR_PTR(-ENOMEM);
 620
 621         /* 8 possible tag values */
 622         tagbits = 0xff;
 623
 624         spin_lock_irqsave(&mns->keys_lock, flags);
 625
 626         /* Walk through the existing keys, looking for potential conflicting
 627          * tags. If we find a conflict, clear that bit from tagbits
 628          */
 629         hlist_for_each_entry(tmp, &mns->keys, hlist) {
 630                 /* We can check the lookup fields (*_addr, tag) without the
 631                  * lock held, they don't change over the lifetime of the key.
 632                  */
 633
 634                 /* if we don't own the tag, it can't conflict */
 635                 if (tmp->tag & MCTP_HDR_FLAG_TO)
 636                         continue;
 637
 638                 if (!(mctp_address_matches(tmp->peer_addr, daddr) &&
 639                       mctp_address_matches(tmp->local_addr, saddr)))
 640                         continue;
 641
 642                 spin_lock(&tmp->lock);
 643                 /* key must still be valid. If we find a match, clear the
 644                  * potential tag value
 645                  */
 646                 if (tmp->valid)
 647                         tagbits &= ~(1 << tmp->tag);
 648                 spin_unlock(&tmp->lock);
 649
 650                 if (!tagbits)
 651                         break;
 652         }
 653
 654         if (tagbits) {
 655                 key->tag = __ffs(tagbits);
 656                 mctp_reserve_tag(net, key, msk);
 657                 trace_mctp_key_acquire(key);
 658
 659                 key->manual_alloc = manual;
 660                 *tagp = key->tag;
 661         }
 662
 663         spin_unlock_irqrestore(&mns->keys_lock, flags);
 664
 665         if (!tagbits) {
 666                 kfree(key);
 667                 return ERR_PTR(-EBUSY);
 668         }
 669
 670         return key;
 671 }
 672
 673 static struct mctp_sk_key *mctp_lookup_prealloc_tag(struct mctp_sock *msk,
 674                                                     mctp_eid_t daddr,
 675                                                     u8 req_tag, u8 *tagp)
 676 {
 677         struct net *net = sock_net(&msk->sk);
 678         struct netns_mctp *mns = &net->mctp;
 679         struct mctp_sk_key *key, *tmp;
 680         unsigned long flags;
 681
 682         req_tag &= ~(MCTP_TAG_PREALLOC | MCTP_TAG_OWNER);
 683         key = NULL;
 684
 685         spin_lock_irqsave(&mns->keys_lock, flags);
 686
 687         hlist_for_each_entry(tmp, &mns->keys, hlist) {
 688                 if (tmp->tag != req_tag)
 689                         continue;
 690
 691                 if (!mctp_address_matches(tmp->peer_addr, daddr))
 692                         continue;
 693
 694                 if (!tmp->manual_alloc)
 695                         continue;
 696
 697                 spin_lock(&tmp->lock);
 698                 if (tmp->valid) {
 699                         key = tmp;
 700                         refcount_inc(&key->refs);
 701                         spin_unlock(&tmp->lock);
 702                         break;
 703                 }
 704                 spin_unlock(&tmp->lock);
 705         }
 706         spin_unlock_irqrestore(&mns->keys_lock, flags);
 707
 708         if (!key)
 709                 return ERR_PTR(-ENOENT);
 710
 711         if (tagp)
 712                 *tagp = key->tag;
 713
 714         return key;
 715 }
 716
 717 /* routing lookups */
 718 static bool mctp_rt_match_eid(struct mctp_route *rt,
 719                               unsigned int net, mctp_eid_t eid)
 720 {
 721         return READ_ONCE(rt->dev->net) == net &&
 722                 rt->min <= eid && rt->max >= eid;
 723 }
 724
 725 /* compares match, used for duplicate prevention */
 726 static bool mctp_rt_compare_exact(struct mctp_route *rt1,
 727                                   struct mctp_route *rt2)
 728 {
 729         ASSERT_RTNL();
 730         return rt1->dev->net == rt2->dev->net &&
 731                 rt1->min == rt2->min &&
 732                 rt1->max == rt2->max;
 733 }
 734
 735 struct mctp_route *mctp_route_lookup(struct net *net, unsigned int dnet,
 736                                      mctp_eid_t daddr)
 737 {
 738         struct mctp_route *tmp, *rt = NULL;
 739
 740         rcu_read_lock();
 741
 742         list_for_each_entry_rcu(tmp, &net->mctp.routes, list) {
 743                 /* TODO: add metrics */
 744                 if (mctp_rt_match_eid(tmp, dnet, daddr)) {
 745                         if (refcount_inc_not_zero(&tmp->refs)) {
 746                                 rt = tmp;
 747                                 break;
 748                         }
 749                 }
 750         }
 751
 752         rcu_read_unlock();
 753
 754         return rt;
 755 }
 756
 757 static struct mctp_route *mctp_route_lookup_null(struct net *net,
 758                                                  struct net_device *dev)
 759 {
 760         struct mctp_route *tmp, *rt = NULL;
 761
 762         rcu_read_lock();
 763
 764         list_for_each_entry_rcu(tmp, &net->mctp.routes, list) {
 765                 if (tmp->dev->dev == dev && tmp->type == RTN_LOCAL &&
 766                     refcount_inc_not_zero(&tmp->refs)) {
 767                         rt = tmp;
 768                         break;
 769                 }
 770         }
 771
 772         rcu_read_unlock();
 773
 774         return rt;
 775 }
 776
 777 static int mctp_do_fragment_route(struct mctp_route *rt, struct sk_buff *skb,
 778                                   unsigned int mtu, u8 tag)
 779 {
 780         const unsigned int hlen = sizeof(struct mctp_hdr);
 781         struct mctp_hdr *hdr, *hdr2;
 782         unsigned int pos, size, headroom;
 783         struct sk_buff *skb2;
 784         int rc;
 785         u8 seq;
 786
 787         hdr = mctp_hdr(skb);
 788         seq = 0;
 789         rc = 0;
 790
 791         if (mtu < hlen + 1) {
 792                 kfree_skb(skb);
 793                 return -EMSGSIZE;
 794         }
 795
 796         /* keep same headroom as the original skb */
 797         headroom = skb_headroom(skb);
 798
 799         /* we've got the header */
 800         skb_pull(skb, hlen);
 801
 802         for (pos = 0; pos < skb->len;) {
 803                 /* size of message payload */
 804                 size = min(mtu - hlen, skb->len - pos);
 805
 806                 skb2 = alloc_skb(headroom + hlen + size, GFP_KERNEL);
 807                 if (!skb2) {
 808                         rc = -ENOMEM;
 809                         break;
 810                 }
 811
 812                 /* generic skb copy */
 813                 skb2->protocol = skb->protocol;
 814                 skb2->priority = skb->priority;
 815                 skb2->dev = skb->dev;
 816                 memcpy(skb2->cb, skb->cb, sizeof(skb2->cb));
 817
 818                 if (skb->sk)
 819                         skb_set_owner_w(skb2, skb->sk);
 820
 821                 /* establish packet */
 822                 skb_reserve(skb2, headroom);
 823                 skb_reset_network_header(skb2);
 824                 skb_put(skb2, hlen + size);
 825                 skb2->transport_header = skb2->network_header + hlen;
 826
 827                 /* copy header fields, calculate SOM/EOM flags & seq */
 828                 hdr2 = mctp_hdr(skb2);
 829                 hdr2->ver = hdr->ver;
 830                 hdr2->dest = hdr->dest;
 831                 hdr2->src = hdr->src;
 832                 hdr2->flags_seq_tag = tag &
 833                         (MCTP_HDR_TAG_MASK | MCTP_HDR_FLAG_TO);
 834
 835                 if (pos == 0)
 836                         hdr2->flags_seq_tag |= MCTP_HDR_FLAG_SOM;
 837
 838                 if (pos + size == skb->len)
 839                         hdr2->flags_seq_tag |= MCTP_HDR_FLAG_EOM;
 840
 841                 hdr2->flags_seq_tag |= seq << MCTP_HDR_SEQ_SHIFT;
 842
 843                 /* copy message payload */
 844                 skb_copy_bits(skb, pos, skb_transport_header(skb2), size);
 845
 846                 /* do route */
 847                 rc = rt->output(rt, skb2);
 848                 if (rc)
 849                         break;
 850
 851                 seq = (seq + 1) & MCTP_HDR_SEQ_MASK;
 852                 pos += size;
 853         }
 854
 855         consume_skb(skb);
 856         return rc;
 857 }
 858
 859 int mctp_local_output(struct sock *sk, struct mctp_route *rt,
 860                       struct sk_buff *skb, mctp_eid_t daddr, u8 req_tag)
 861 {
 862         struct mctp_sock *msk = container_of(sk, struct mctp_sock, sk);
 863         struct mctp_skb_cb *cb = mctp_cb(skb);
 864         struct mctp_route tmp_rt = {0};
 865         struct mctp_sk_key *key;
 866         struct mctp_hdr *hdr;
 867         unsigned long flags;
 868         unsigned int mtu;
 869         mctp_eid_t saddr;
 870         bool ext_rt;
 871         int rc;
 872         u8 tag;
 873
 874         rc = -ENODEV;
 875
 876         if (rt) {
 877                 ext_rt = false;
 878                 if (WARN_ON(!rt->dev))
 879                         goto out_release;
 880
 881         } else if (cb->ifindex) {
 882                 struct net_device *dev;
 883
 884                 ext_rt = true;
 885                 rt = &tmp_rt;
 886
 887                 rcu_read_lock();
 888                 dev = dev_get_by_index_rcu(sock_net(sk), cb->ifindex);
 889                 if (!dev) {
 890                         rcu_read_unlock();
 891                         return rc;
 892                 }
 893                 rt->dev = __mctp_dev_get(dev);
 894                 rcu_read_unlock();
 895
 896                 if (!rt->dev)
 897                         goto out_release;
 898
 899                 /* establish temporary route - we set up enough to keep
 900                  * mctp_route_output happy
 901                  */
 902                 rt->output = mctp_route_output;
 903                 rt->mtu = 0;
 904
 905         } else {
 906                 return -EINVAL;
 907         }
 908
 909         spin_lock_irqsave(&rt->dev->addrs_lock, flags);
 910         if (rt->dev->num_addrs == 0) {
 911                 rc = -EHOSTUNREACH;
 912         } else {
 913                 /* use the outbound interface's first address as our source */
 914                 saddr = rt->dev->addrs[0];
 915                 rc = 0;
 916         }
 917         spin_unlock_irqrestore(&rt->dev->addrs_lock, flags);
 918
 919         if (rc)
 920                 goto out_release;
 921
 922         if (req_tag & MCTP_TAG_OWNER) {
 923                 if (req_tag & MCTP_TAG_PREALLOC)
 924                         key = mctp_lookup_prealloc_tag(msk, daddr,
 925                                                        req_tag, &tag);
 926                 else
 927                         key = mctp_alloc_local_tag(msk, daddr, saddr,
 928                                                    false, &tag);
 929
 930                 if (IS_ERR(key)) {
 931                         rc = PTR_ERR(key);
 932                         goto out_release;
 933                 }
 934                 mctp_skb_set_flow(skb, key);
 935                 /* done with the key in this scope */
 936                 mctp_key_unref(key);
 937                 tag |= MCTP_HDR_FLAG_TO;
 938         } else {
 939                 key = NULL;
 940                 tag = req_tag & MCTP_TAG_MASK;
 941         }
 942
 943         skb->protocol = htons(ETH_P_MCTP);
 944         skb->priority = 0;
 945         skb_reset_transport_header(skb);
 946         skb_push(skb, sizeof(struct mctp_hdr));
 947         skb_reset_network_header(skb);
 948         skb->dev = rt->dev->dev;
 949
 950         /* cb->net will have been set on initial ingress */
 951         cb->src = saddr;
 952
 953         /* set up common header fields */
 954         hdr = mctp_hdr(skb);
 955         hdr->ver = 1;
 956         hdr->dest = daddr;
 957         hdr->src = saddr;
 958
 959         mtu = mctp_route_mtu(rt);
 960
 961         if (skb->len + sizeof(struct mctp_hdr) <= mtu) {
 962                 hdr->flags_seq_tag = MCTP_HDR_FLAG_SOM |
 963                         MCTP_HDR_FLAG_EOM | tag;
 964                 rc = rt->output(rt, skb);
 965         } else {
 966                 rc = mctp_do_fragment_route(rt, skb, mtu, tag);
 967         }
 968
 969 out_release:
 970         if (!ext_rt)
 971                 mctp_route_release(rt);
 972
 973         mctp_dev_put(tmp_rt.dev);
 974
 975         return rc;
 976 }
 977
 978 /* route management */
 979 static int mctp_route_add(struct mctp_dev *mdev, mctp_eid_t daddr_start,
 980                           unsigned int daddr_extent, unsigned int mtu,
 981                           unsigned char type)
 982 {
 983         int (*rtfn)(struct mctp_route *rt, struct sk_buff *skb);
 984         struct net *net = dev_net(mdev->dev);
 985         struct mctp_route *rt, *ert;
 986
 987         if (!mctp_address_unicast(daddr_start))
 988                 return -EINVAL;
 989
 990         if (daddr_extent > 0xff || daddr_start + daddr_extent >= 255)
 991                 return -EINVAL;
 992
 993         switch (type) {
 994         case RTN_LOCAL:
 995                 rtfn = mctp_route_input;
 996                 break;
 997         case RTN_UNICAST:
 998                 rtfn = mctp_route_output;
 999                 break;
1000         default:
1001                 return -EINVAL;
1002         }
1003
1004         rt = mctp_route_alloc();
1005         if (!rt)
1006                 return -ENOMEM;
1007
1008         rt->min = daddr_start;
1009         rt->max = daddr_start + daddr_extent;
1010         rt->mtu = mtu;
1011         rt->dev = mdev;
1012         mctp_dev_hold(rt->dev);
1013         rt->type = type;
1014         rt->output = rtfn;
1015
1016         ASSERT_RTNL();
1017         /* Prevent duplicate identical routes. */
1018         list_for_each_entry(ert, &net->mctp.routes, list) {
1019                 if (mctp_rt_compare_exact(rt, ert)) {
1020                         mctp_route_release(rt);
1021                         return -EEXIST;
1022                 }
1023         }
1024
1025         list_add_rcu(&rt->list, &net->mctp.routes);
1026
1027         return 0;
1028 }
1029
1030 static int mctp_route_remove(struct mctp_dev *mdev, mctp_eid_t daddr_start,
1031                              unsigned int daddr_extent, unsigned char type)
1032 {
1033         struct net *net = dev_net(mdev->dev);
1034         struct mctp_route *rt, *tmp;
1035         mctp_eid_t daddr_end;
1036         bool dropped;
1037
1038         if (daddr_extent > 0xff || daddr_start + daddr_extent >= 255)
1039                 return -EINVAL;
1040
1041         daddr_end = daddr_start + daddr_extent;
1042         dropped = false;
1043
1044         ASSERT_RTNL();
1045
1046         list_for_each_entry_safe(rt, tmp, &net->mctp.routes, list) {
1047                 if (rt->dev == mdev &&
1048                     rt->min == daddr_start && rt->max == daddr_end &&
1049                     rt->type == type) {
1050                         list_del_rcu(&rt->list);
1051                         /* TODO: immediate RTM_DELROUTE */
1052                         mctp_route_release(rt);
1053                         dropped = true;
1054                 }
1055         }
1056
1057         return dropped ? 0 : -ENOENT;
1058 }
1059
1060 int mctp_route_add_local(struct mctp_dev *mdev, mctp_eid_t addr)
1061 {
1062         return mctp_route_add(mdev, addr, 0, 0, RTN_LOCAL);
1063 }
1064
1065 int mctp_route_remove_local(struct mctp_dev *mdev, mctp_eid_t addr)
1066 {
1067         return mctp_route_remove(mdev, addr, 0, RTN_LOCAL);
1068 }
1069
1070 /* removes all entries for a given device */
1071 void mctp_route_remove_dev(struct mctp_dev *mdev)
1072 {
1073         struct net *net = dev_net(mdev->dev);
1074         struct mctp_route *rt, *tmp;
1075
1076         ASSERT_RTNL();
1077         list_for_each_entry_safe(rt, tmp, &net->mctp.routes, list) {
1078                 if (rt->dev == mdev) {
1079                         list_del_rcu(&rt->list);
1080                         /* TODO: immediate RTM_DELROUTE */
1081                         mctp_route_release(rt);
1082                 }
1083         }
1084 }
1085
1086 /* Incoming packet-handling */
1087
1088 static int mctp_pkttype_receive(struct sk_buff *skb, struct net_device *dev,
1089                                 struct packet_type *pt,
1090                                 struct net_device *orig_dev)
1091 {
1092         struct net *net = dev_net(dev);
1093         struct mctp_dev *mdev;
1094         struct mctp_skb_cb *cb;
1095         struct mctp_route *rt;
1096         struct mctp_hdr *mh;
1097
1098         rcu_read_lock();
1099         mdev = __mctp_dev_get(dev);
1100         rcu_read_unlock();
1101         if (!mdev) {
1102                 /* basic non-data sanity checks */
1103                 goto err_drop;
1104         }
1105
1106         if (!pskb_may_pull(skb, sizeof(struct mctp_hdr)))
1107                 goto err_drop;
1108
1109         skb_reset_transport_header(skb);
1110         skb_reset_network_header(skb);
1111
1112         /* We have enough for a header; decode and route */
1113         mh = mctp_hdr(skb);
1114         if (mh->ver < MCTP_VER_MIN || mh->ver > MCTP_VER_MAX)
1115                 goto err_drop;
1116
1117         /* source must be valid unicast or null; drop reserved ranges and
1118          * broadcast
1119          */
1120         if (!(mctp_address_unicast(mh->src) || mctp_address_null(mh->src)))
1121                 goto err_drop;
1122
1123         /* dest address: as above, but allow broadcast */
1124         if (!(mctp_address_unicast(mh->dest) || mctp_address_null(mh->dest) ||
1125               mctp_address_broadcast(mh->dest)))
1126                 goto err_drop;
1127
1128         /* MCTP drivers must populate halen/haddr */
1129         if (dev->type == ARPHRD_MCTP) {
1130                 cb = mctp_cb(skb);
1131         } else {
1132                 cb = __mctp_cb(skb);
1133                 cb->halen = 0;
1134         }
1135         cb->net = READ_ONCE(mdev->net);
1136         cb->ifindex = dev->ifindex;
1137
1138         rt = mctp_route_lookup(net, cb->net, mh->dest);
1139
1140         /* NULL EID, but addressed to our physical address */
1141         if (!rt && mh->dest == MCTP_ADDR_NULL && skb->pkt_type == PACKET_HOST)
1142                 rt = mctp_route_lookup_null(net, dev);
1143
1144         if (!rt)
1145                 goto err_drop;
1146
1147         rt->output(rt, skb);
1148         mctp_route_release(rt);
1149         mctp_dev_put(mdev);
1150
1151         return NET_RX_SUCCESS;
1152
1153 err_drop:
1154         kfree_skb(skb);
1155         mctp_dev_put(mdev);
1156         return NET_RX_DROP;
1157 }
1158
1159 static struct packet_type mctp_packet_type = {
1160         .type = cpu_to_be16(ETH_P_MCTP),
1161         .func = mctp_pkttype_receive,
1162 };
1163
1164 /* netlink interface */
1165
1166 static const struct nla_policy rta_mctp_policy[RTA_MAX + 1] = {
1167         [RTA_DST]               = { .type = NLA_U8 },
1168         [RTA_METRICS]           = { .type = NLA_NESTED },
1169         [RTA_OIF]               = { .type = NLA_U32 },
1170 };
1171
1172 /* Common part for RTM_NEWROUTE and RTM_DELROUTE parsing.
1173  * tb must hold RTA_MAX+1 elements.
1174  */
1175 static int mctp_route_nlparse(struct sk_buff *skb, struct nlmsghdr *nlh,
1176                               struct netlink_ext_ack *extack,
1177                               struct nlattr **tb, struct rtmsg **rtm,
1178                               struct mctp_dev **mdev, mctp_eid_t *daddr_start)
1179 {
1180         struct net *net = sock_net(skb->sk);
1181         struct net_device *dev;
1182         unsigned int ifindex;
1183         int rc;
1184
1185         rc = nlmsg_parse(nlh, sizeof(struct rtmsg), tb, RTA_MAX,
1186                          rta_mctp_policy, extack);
1187         if (rc < 0) {
1188                 NL_SET_ERR_MSG(extack, "incorrect format");
1189                 return rc;
1190         }
1191
1192         if (!tb[RTA_DST]) {
1193                 NL_SET_ERR_MSG(extack, "dst EID missing");
1194                 return -EINVAL;
1195         }
1196         *daddr_start = nla_get_u8(tb[RTA_DST]);
1197
1198         if (!tb[RTA_OIF]) {
1199                 NL_SET_ERR_MSG(extack, "ifindex missing");
1200                 return -EINVAL;
1201         }
1202         ifindex = nla_get_u32(tb[RTA_OIF]);
1203
1204         *rtm = nlmsg_data(nlh);
1205         if ((*rtm)->rtm_family != AF_MCTP) {
1206                 NL_SET_ERR_MSG(extack, "route family must be AF_MCTP");
1207                 return -EINVAL;
1208         }
1209
1210         dev = __dev_get_by_index(net, ifindex);
1211         if (!dev) {
1212                 NL_SET_ERR_MSG(extack, "bad ifindex");
1213                 return -ENODEV;
1214         }
1215         *mdev = mctp_dev_get_rtnl(dev);
1216         if (!*mdev)
1217                 return -ENODEV;
1218
1219         if (dev->flags & IFF_LOOPBACK) {
1220                 NL_SET_ERR_MSG(extack, "no routes to loopback");
1221                 return -EINVAL;
1222         }
1223
1224         return 0;
1225 }
1226
1227 static const struct nla_policy rta_metrics_policy[RTAX_MAX + 1] = {
1228         [RTAX_MTU]              = { .type = NLA_U32 },
1229 };
1230
1231 static int mctp_newroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1232                          struct netlink_ext_ack *extack)
1233 {
1234         struct nlattr *tb[RTA_MAX + 1];
1235         struct nlattr *tbx[RTAX_MAX + 1];
1236         mctp_eid_t daddr_start;
1237         struct mctp_dev *mdev;
1238         struct rtmsg *rtm;
1239         unsigned int mtu;
1240         int rc;
1241
1242         rc = mctp_route_nlparse(skb, nlh, extack, tb,
1243                                 &rtm, &mdev, &daddr_start);
1244         if (rc < 0)
1245                 return rc;
1246
1247         if (rtm->rtm_type != RTN_UNICAST) {
1248                 NL_SET_ERR_MSG(extack, "rtm_type must be RTN_UNICAST");
1249                 return -EINVAL;
1250         }
1251
1252         mtu = 0;
1253         if (tb[RTA_METRICS]) {
1254                 rc = nla_parse_nested(tbx, RTAX_MAX, tb[RTA_METRICS],
1255                                       rta_metrics_policy, NULL);
1256                 if (rc < 0)
1257                         return rc;
1258                 if (tbx[RTAX_MTU])
1259                         mtu = nla_get_u32(tbx[RTAX_MTU]);
1260         }
1261
1262         rc = mctp_route_add(mdev, daddr_start, rtm->rtm_dst_len, mtu,
1263                             rtm->rtm_type);
1264         return rc;
1265 }
1266
1267 static int mctp_delroute(struct sk_buff *skb, struct nlmsghdr *nlh,
1268                          struct netlink_ext_ack *extack)
1269 {
1270         struct nlattr *tb[RTA_MAX + 1];
1271         mctp_eid_t daddr_start;
1272         struct mctp_dev *mdev;
1273         struct rtmsg *rtm;
1274         int rc;
1275
1276         rc = mctp_route_nlparse(skb, nlh, extack, tb,
1277                                 &rtm, &mdev, &daddr_start);
1278         if (rc < 0)
1279                 return rc;
1280
1281         /* we only have unicast routes */
1282         if (rtm->rtm_type != RTN_UNICAST)
1283                 return -EINVAL;
1284
1285         rc = mctp_route_remove(mdev, daddr_start, rtm->rtm_dst_len, RTN_UNICAST);
1286         return rc;
1287 }
1288
1289 static int mctp_fill_rtinfo(struct sk_buff *skb, struct mctp_route *rt,
1290                             u32 portid, u32 seq, int event, unsigned int flags)
1291 {
1292         struct nlmsghdr *nlh;
1293         struct rtmsg *hdr;
1294         void *metrics;
1295
1296         nlh = nlmsg_put(skb, portid, seq, event, sizeof(*hdr), flags);
1297         if (!nlh)
1298                 return -EMSGSIZE;
1299
1300         hdr = nlmsg_data(nlh);
1301         hdr->rtm_family = AF_MCTP;
1302
1303         /* we use the _len fields as a number of EIDs, rather than
1304          * a number of bits in the address
1305          */
1306         hdr->rtm_dst_len = rt->max - rt->min;
1307         hdr->rtm_src_len = 0;
1308         hdr->rtm_tos = 0;
1309         hdr->rtm_table = RT_TABLE_DEFAULT;
1310         hdr->rtm_protocol = RTPROT_STATIC; /* everything is user-defined */
1311         hdr->rtm_scope = RT_SCOPE_LINK; /* TODO: scope in mctp_route? */
1312         hdr->rtm_type = rt->type;
1313
1314         if (nla_put_u8(skb, RTA_DST, rt->min))
1315                 goto cancel;
1316
1317         metrics = nla_nest_start_noflag(skb, RTA_METRICS);
1318         if (!metrics)
1319                 goto cancel;
1320
1321         if (rt->mtu) {
1322                 if (nla_put_u32(skb, RTAX_MTU, rt->mtu))
1323                         goto cancel;
1324         }
1325
1326         nla_nest_end(skb, metrics);
1327
1328         if (rt->dev) {
1329                 if (nla_put_u32(skb, RTA_OIF, rt->dev->dev->ifindex))
1330                         goto cancel;
1331         }
1332
1333         /* TODO: conditional neighbour physaddr? */
1334
1335         nlmsg_end(skb, nlh);
1336
1337         return 0;
1338
1339 cancel:
1340         nlmsg_cancel(skb, nlh);
1341         return -EMSGSIZE;
1342 }
1343
1344 static int mctp_dump_rtinfo(struct sk_buff *skb, struct netlink_callback *cb)
1345 {
1346         struct net *net = sock_net(skb->sk);
1347         struct mctp_route *rt;
1348         int s_idx, idx;
1349
1350         /* TODO: allow filtering on route data, possibly under
1351          * cb->strict_check
1352          */
1353
1354         /* TODO: change to struct overlay */
1355         s_idx = cb->args[0];
1356         idx = 0;
1357
1358         rcu_read_lock();
1359         list_for_each_entry_rcu(rt, &net->mctp.routes, list) {
1360                 if (idx++ < s_idx)
1361                         continue;
1362                 if (mctp_fill_rtinfo(skb, rt,
1363                                      NETLINK_CB(cb->skb).portid,
1364                                      cb->nlh->nlmsg_seq,
1365                                      RTM_NEWROUTE, NLM_F_MULTI) < 0)
1366                         break;
1367         }
1368
1369         rcu_read_unlock();
1370         cb->args[0] = idx;
1371
1372         return skb->len;
1373 }
1374
1375 /* net namespace implementation */
1376 static int __net_init mctp_routes_net_init(struct net *net)
1377 {
1378         struct netns_mctp *ns = &net->mctp;
1379
1380         INIT_LIST_HEAD(&ns->routes);
1381         INIT_HLIST_HEAD(&ns->binds);
1382         mutex_init(&ns->bind_lock);
1383         INIT_HLIST_HEAD(&ns->keys);
1384         spin_lock_init(&ns->keys_lock);
1385         WARN_ON(mctp_default_net_set(net, MCTP_INITIAL_DEFAULT_NET));
1386         return 0;
1387 }
1388
1389 static void __net_exit mctp_routes_net_exit(struct net *net)
1390 {
1391         struct mctp_route *rt;
1392
1393         rcu_read_lock();
1394         list_for_each_entry_rcu(rt, &net->mctp.routes, list)
1395                 mctp_route_release(rt);
1396         rcu_read_unlock();
1397 }
1398
1399 static struct pernet_operations mctp_net_ops = {
1400         .init = mctp_routes_net_init,
1401         .exit = mctp_routes_net_exit,
1402 };
1403
1404 int __init mctp_routes_init(void)
1405 {
1406         dev_add_pack(&mctp_packet_type);
1407
1408         rtnl_register_module(THIS_MODULE, PF_MCTP, RTM_GETROUTE,
1409                              NULL, mctp_dump_rtinfo, 0);
1410         rtnl_register_module(THIS_MODULE, PF_MCTP, RTM_NEWROUTE,
1411                              mctp_newroute, NULL, 0);
1412         rtnl_register_module(THIS_MODULE, PF_MCTP, RTM_DELROUTE,
1413                              mctp_delroute, NULL, 0);
1414
1415         return register_pernet_subsys(&mctp_net_ops);
1416 }
1417
1418 void mctp_routes_exit(void)
1419 {
1420         unregister_pernet_subsys(&mctp_net_ops);
1421         rtnl_unregister(PF_MCTP, RTM_DELROUTE);
1422         rtnl_unregister(PF_MCTP, RTM_NEWROUTE);
1423         rtnl_unregister(PF_MCTP, RTM_GETROUTE);
1424         dev_remove_pack(&mctp_packet_type);
1425 }
1426
1427 #if IS_ENABLED(CONFIG_MCTP_TEST)
1428 #include "test/route-test.c"
1429 #endif